Codec with filtering and / or prediction processing of a precision reduced spectral domain representation and / or with prediction processing using a prediction information

Abstract

Embodiments comprise a decoder for decoding a signal, wherein the decoder is configured to decode an encoded representation of a precision reduced spectral domain representation of a signal, in order to obtain a decoded version of the precision reduced spectral domain representation, to decode an encoded representation of a residual information, in order to obtain a decoded version of the residual information, to obtain a filtered and / or prediction processed precision reduced spectral domain representation on the basis of the precision reduced spectral domain representation using a filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and / or using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and to determine a combined spectrum based on the filtered and / or prediction processed precision reduced spectral domain representation and the decoded version of the residual information. Furthermore, encoders, methods for decoding, methods for encoding, computer programs and bitstreams are disclosed.

Description

[0001] Decoder, Encoder, Method, Computer Program and Bitstream with Filtering and / or Prediction Processing of a precision reduced Spectral Domain Representation and / or with Prediction Processing using a Prediction Information

[0002] Technical Field

[0003] Embodiments according to the invention comprise decoders, encoders, methods, computer programs and bitstreams with filtering and / or with prediction processing of a precision reduced spectral domain representation.

[0004] Embodiments according to the invention comprise decoders, encoders, methods, computer programs and bitstreams for temporal noise shaping, TNS, for (e.g. near) lossless data coding, e.g. audio coding.

[0005] Background of the Invention

[0006] Temporal Noise Shaping is a well-known coding tool and for instance used in MPEG-4 AAC- LC or ETSI LC3plus. The basic concept is described in [1] and [2],

[0007] Basically, a prediction filter in frequency domain on the encoder side may remove energy from the signal which may, at least partially, be restored again on the decoder side by the inverse filter. Any quantization error added in between may be shaped temporally by the decoder side inverse filter. All TNS implementations so far focus on the noise shaping aspect.

[0008] However, it was recognized that such filtering or prediction approaches may as well have undesired effects, which may lead to increased errors in coded data, for example in particular for near losslessly encoded signals.

[0009] Hence, there is a desire to obtain a concept for a coding of data, such as audio coding, which achieves an improved compromise between a computational effort, a quality, e.g. in particular perceptual quality, of the coded data, and a scalability, e.g. with regard to available bitrates.

[0010] This is achieved by the subject matter of the independent claims of the present application.

[0011] Further embodiments according to the invention are defined by the subject matter of the dependent claims of the present application.

[0012] FV - ACr - FH240905PEP-2025315045. DOCX of the invention

[0013] Embodiments according to the invention comprise a decoder for decoding a signal, wherein the decoder is configured to decode an encoded representation of a precision reduced spectral domain representation of a signal (e.g. a filtered and / or prediction processed version of a precision reduced spectral domain representation, e.g. as provided in the encoder; e.g. a TNS filtered spectrum with reduced precision; e.g. a more significant spectral value information, e.g. from a bitstream), in order to obtain a decoded version of the precision reduced spectral domain representation. Furthermore, the decoder is configured to decode an encoded representation of a residual information (e.g. a residual spectral domain representation or a portion thereof, e.g. a residual between a spectral domain representation and the precision reduced spectral domain representation; which is, for example, a residual of the precision reduced spectral domain representation; which may, for example, represent a difference between the spectral domain representation and the precision reduced spectral domain representation; e.g. a residual of reduced spectrum; e.g. a residual between the spectral domain representation and the precision reduced spectral domain representation; e.g. a less significant spectral value information, e.g. corresponding to the more significant spectral value information; e.g. from a bitstream), in order to obtain a decoded version of the residual information.

[0014] Furthermore, the decoder is configured to obtain a filtered and / or prediction processed precision reduced spectral domain representation (e.g. the precision reduced spectral domain representation) on the basis of the precision reduced spectral domain representation using a filtering (e.g. an inverse filtering; e.g. an inverse prediction filtering; e.g. a prediction filtering) in a frequency direction (e.g. over frequency; e.g. across frequency; e.g. using a frequency domain prediction filtering, e.g. using a frequency domain inverse prediction filtering; e.g. using a filtering of the precision reduced spectral domain representation using a TNS filter, e.g. using an inverse TNS filter; e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value) (wherein, for example, the encoder is configured to apply a prediction filter, e.g. an inverse prediction filter, operating in the frequency direction to the filtered precision reduced spectral domain representation, to obtain the precision reduced spectral domain representation) (wherein, for example, the encoder is configured to filter the filtered precision reduced spectral domain representation using a prediction filter operating in the frequency direction, to obtain the precision reduced

[0015] FH240905PEP-2025315045. DOCX spectral domain representation) (wherein, for example, the encoder is configured to apply a prediction filtering, e.g. an inverse prediction filtering, to the filtered precision reduced spectral domain representation, in order to obtain a precision reduced spectral domain representation) (e.g. apply an inverse prediction filtering to the decoded version of the prediction filtered precision reduced spectral domain representation, in order to obtain the precision reduced spectral domain representation) and / or in a time direction and / or in a spatial direction and / or in a channel direction.

[0016] Alternatively or in addition, the decoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation on the basis of the precision reduced spectral domain representation using a prediction processing (e.g. a decoder-sided prediction processing, e.g. a prediction, e.g. a reconstruction of previously, encoder-sided reduced, redundant information) in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction.

[0017] Furthermore, the decoder is configured to determine a combined spectrum based on the filtered and / or prediction processed precision reduced spectral domain representation and the decoded version of the residual information.

[0018] Embodiments according to the invention comprise an encoder for encoding a signal (e.g. an audio signal), wherein the encoder is configured to obtain (e.g. derive, e.g. receive) a spectral domain representation (e.g. an initial spectral domain representation; e.g. an original precision spectral domain representation; e.g. a spectral domain representation in a full integer value representation; e.g. an MDCT spectrum) of the signal.

[0019] Furthermore, the encoder is configured to obtain (e.g. determine, e.g. to calculate) a precision reduced spectral domain representation (e.g. a precision reduced spectrum) of the signal based on the spectral domain representation.

[0020] Furthermore, the encoder is configured to obtain (e.g. determine, e.g. to calculate, e.g. to approximate, e.g. to set, e.g. to estimate) a residual information (e.g. a residual spectral domain representation or a portion thereof, e.g. a residual between the spectral domain representation and the precision reduced spectral domain representation; which is, for example, a residual of the precision reduced spectral domain representation; which may, for example, represent a difference between the spectral domain representation and the precision reduced spectral domain representation; e.g. a residual of reduced spectrum) based on (e.g. on the basis of; e.g. in dependence on) the spectral domain representation (wherein the

[0021] FH240905PEP-2025315045. DOCX encoder may, for example, be configured to split up the spectral domain representation of the signal into the precision reduced spectral domain representation and a residual of the precision reduced spectral domain representation, in order to obtain the precision reduced spectral domain representation and the residual information).

[0022] Furthermore, the encoder is configured to obtain a filtered and / or prediction processed precision reduced spectral domain representation (e.g. a filtered version of the precision reduced spectral domain representation; e.g. a TNS filtered spectrum with reduced precision) on the basis of the precision reduced spectral domain representation using a filtering (e.g. a prediction filtering) in a frequency direction (e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering; e.g. using a filtering of the precision reduced spectral domain representation using a TNS filter; e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value; wherein, for example, the residual spectral domain representation is left out of consideration when obtaining the filtered precision reduced spectral domain representation) (wherein, for example, the encoder is configured to apply a prediction filter operating in the frequency direction to the precision reduced spectral domain representation (but not to the residual spectral domain representation), to obtain the filtered precision reduced spectral domain representation; wherein, for example, the encoder is configured to filter the precision reduced spectral domain representation (but not the residual spectral domain representation) using a prediction filter operating in the frequency direction, to obtain the filtered precision reduced spectral domain representation) (wherein, for example, the encoder is configured to apply a prediction filtering to the precision reduced spectral domain representation, in order to obtain a prediction filtered precision reduced spectral domain representation) and / or in a time direction and / or in a spatial direction and / or in a channel direction.

[0023] Alternatively or in addition, the encoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation on the basis of the precision reduced spectral domain representation using a prediction processing (e.g. an encoder-sided prediction, e.g. an inverse prediction, e.g. a redundancy reduction) in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction.

[0024] Furthermore, the encoder is configured to obtain an encoded representation on the basis of the filtered and / or prediction processed precision reduced spectral domain representation and the residual information (wherein, for example, the encoder is configured to encode the filtered

[0025] FH240905PEP-2025315045. DOCX precision reduced spectral domain representation, or a processed version of the precision reduced spectral domain representation, or a part of the precision reduced spectral domain representation, and wherein, for example, the encoder is configured to encode the residual information, or a processed version of the residual information, or a part of the residual information (e.g. as an index to a table, e.g. a look-up table, comprising approximations or default values for residuals; e.g. in a bitstream)).

[0026] It was recognized that limiting a filtering and / or prediction processing step in a coding procedure of a spectral domain representation of a signal to a precision reduced spectral domain representation of the signal (so that a residual of the spectral domain representation is not prediction processed and / or filtered) enables reducing an error of the coded spectral data, for example, in particular for a near lossless coding.

[0027] In this regard, it is to be noted, regarding a full encoding-decoding-processing chain, that the encoder-sided filtering and / or prediction processing and decoder-sided filtering and / or prediction processing may, for example, be considered (or for example intended) as inverse operations of each other. For example, encoder and decoder may be configured to perform corresponding inverse filterings and / or inverse prediction processings (e.g. such as an encoder-sided redundancy reduction and a decoder-sided reconstruction of the removed, redundant information).

[0028] For example, this may be performed such that the original representation (e.g. the spectral domain representation of the signal) can be reconstructed on the decoder side whereas the filtered and / or prediction processed representation transmitted between encoder or decoder can be transmitted with lower bitrate than the original representation.

[0029] Hence, the encoder-sided prediction processing may, for example, be an inverse prediction processing; e.g., a prediction processing removing predictable signal portions; e.g., a prediction processing determining a prediction residual or excitation signal such that using a (e.g., decoder-sided) corresponding (e.g., forward instead of inverse) prediction processing on the basis of the prediction residual or excitation signal the precision reduced spectral domain representation may be reconstructed.

[0030] For example, in a lossy operating mode (e.g. a noise shaping operating mode), a filtering and / or prediction processing (e.g. comprising no rounding) of a spectral domain representation of a signal may enable a perception-based quantization noise shaping. For example, in a lossless operating mode (e.g. redundancy reduction operating mode), the same or similar

[0031] FH240905PEP-2025315045. DOCX filtering and / or prediction processing (e.g. comprising additionally a rounding) of the spectral domain representation of the signal (e.g. an integer domain representation) may achieve reducing (and respectively reinstating) the energy of the spectral domain representation, e.g. reducing redundancy, hence enabling high prediction gains.

[0032] However, it was recognized that such a filtering and / or prediction processing of the spectral domain representation of the signal may, for example, alter an amplitude of the signal significantly. It was recognized that if the transmission of the filtered and / or prediction processed version of the spectral domain representation is lossy (e.g. dropping some bits, e.g. even only some least significant bits; e.g. introducing a significant quantization error) a mean square error of the coded spectral data may increase significantly, e.g. in particular for filterings with significant gains. Hence, artifacts may be introduced in the signal, which may be particularly problematic in a near lossless operating mode (which may comprise filter parametrizations achieving high prediction gains).

[0033] The inventive decomposition of the spectral domain representation into a precision reduced version thereof and a residual enables creating well defined constraints for how the filtering and / or prediction processing affects the precision reduced version of the spectral domain representation.

[0034] In particular, the precision reduced version of the spectral domain representation may be obtained so that its filtered and / or prediction processed version can be fully, e.g. losslessly, transmitted, e.g. given an available bitrate. As an example, hence, a quantization error may not be introduced in this signal portion such that a filtering and / or prediction processing may not introduce an error in this signal portion. The residual information may bypass the filtering / prediction processing step and may hence, although not necessarily transmitted fully (e.g. because of an available bitrate) not be subject to filter-based alterations of quantization noise.

[0035] The precision reduced version of the spectral domain representation may, in particular, comprise a more significant portion of information of the spectral domain representation, such as most significant bits.

[0036] For example, based on an available bitrate, a selective allocation of some bits of the spectral domain representation to the precision reduced version of the spectral domain representation and of remaining bits of the spectral domain representation to the residual information may be

[0037] FH240905PEP-2025315045. DOCX performed. This may enable “dropping” less important bits of the residual to satisfy bitrate constraints.

[0038] In other words, embodiments enable creating a lossless signal transmission with filtering and / or prediction processing for the precision reduced version of the spectral domain representation, so that the filtering and / or prediction processing operations in encoder and decoder are inverted versions of one another and a scalable (e.g. potentially lossy or lossless) transmission without filtering and / or prediction processing for the residual.

[0039] Furthermore, this inventive approach enables controlling how a decline in available bitrate affects the transmitted signal in a very precise manner.

[0040] In other words, and as an example, based on an available bitrate, the extent to which the filtering and / or prediction processing is applied to the spectral domain representation may be controlled, e.g. by controlling how the spectral domain representation is dismantled into the precision reduced version of the spectral domain representation and the residual information and by separating the processing of the precision reduced spectral domain representation and of the residual information.

[0041] With respect to the above-discussed embodiments, it is to be noted that, as an optional feature, the filtering and / or prediction processing (e.g. decoder-sided and for example in a corresponding, e.g. inverse, manner encoder-sided) may, for example, comprise or may, for example, be implemented as or may, for example, be one or more of the following: Interchannel prediction, Joint stereo coding, Mid / side coding, Complex stereo prediction, Multichannel coding tool; Long Term Prediction; Harmonic Prediction; Temporal Noise Shaping; Frequency Domain Prediction.

[0042] Hence, encoders (e.g. audio encoders) according to embodiments may be configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (e.g. in the form of a prediction processed precision reduced spectral domain representation) using one or more of: Inter-channel prediction, Joint stereo coding, Mid / side coding, Complex stereo prediction, Multi-channel coding tool; Long Term Prediction; Harmonic Prediction; Temporal Noise Shaping; Frequency Domain Prediction.

[0043] Hence, decoders (e.g. audio decoders) according to embodiments may be configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (e.g. in the form of a prediction processed precision reduced spectral domain representation) using

[0044] FH240905PEP-2025315045. DOCX one or more of: Inter-channel prediction, Joint stereo coding, Mid / side coding, Complex stereo prediction, Multi-channel coding tool; Long Term Prediction; Harmonic Prediction; Temporal Noise Shaping; Frequency Domain Prediction.

[0045] Again, it is to be noted that a respective decoder-sided filtering and / or prediction processing may correspond to an encoder-sided “inverse” filtering and / or prediction processing.

[0046] Hence, as an example, a prediction processing in a frequency direction, and / or in a time direction and / or in a spatial direction and / or in a channel direction may be achieved, depending on the specific prediction or coding approach (or combinations thereof).

[0047] Hence, in other words, examples for a prediction processing or prediction, according to embodiments, e.g. in particular for audio coding, may comprise one or more of: Inter-channel prediction, Joint stereo coding, Mid / side coding, Complex stereo prediction, Multi-channel coding tool; Long Term Prediction; Harmonic Prediction; Temporal Noise Shaping; Frequency Domain Prediction.

[0048] Embodiments according to the invention comprise an encoder, for encoding a signal (e.g. an audio signal), wherein the encoder is configured to switch between a noise shaping operating mode (e.g. a TNS operating mode; e.g. an operating mode in which the coefficients of the filter structure are determined such that the filter effects a temporal noise shaping, e.g. a temporal noise shaping in which a quantization noise of a decoded signal is reduced before a transient, or a temporal noise shaping in which a temporal evolution of the quantization noise of a decoded signal is adapted to (e.g. approximately follows) a signal envelope; e.g. a lossy mode and / or a near lossless mode according to a Fallback strategy), and a redundancy reduction operating mode (e.g. a prediction gain operating mode; e.g. a lossless mode and / or a near lossless mode according to an expected case).

[0049] Furthermore, the encoder is configured to obtain (e.g. derive, e.g. receive) a spectral domain representation of the signal (e.g. an initial spectral domain representation; e.g. an original precision spectral domain representation; e.g. a spectral domain representation in a full integer value representation; e.g. an MDCT spectrum).

[0050] Furthermore, the encoder is configured to (e.g. in the TNS operating mode) obtain a filtered and / or prediction processed spectral domain representation (e.g. a filtered version of the spectral domain representation; e.g. a TNS filtered spectrum) on the basis of the spectral domain representation using a filter structure (e.g. a TNS filter structure; e.g. a TNS filter

[0051] FH240905PEP-2025315045. DOCX module implemented using software or using hardware, or using a combination of software and hardware) (e.g. using a filtering (e.g. a prediction filtering) in a frequency direction (e.g. a prediction filtering adapted to perform a noise shaping; e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering; e.g. using a filtering of the spectral domain representation using a TNS filter; e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value; wherein, for example, no residual spectral domain representation is determined or considered when obtaining the filtered spectral domain representation) (wherein, for example, the encoder is configured to apply a prediction filter operating in the frequency direction to the spectral domain representation, to obtain the filtered spectral domain representation) (wherein, for example, the encoder is configured to filter the spectral domain representation using a prediction filter operating in the frequency direction, to obtain the filtered spectral domain representation; wherein, for example, the encoder is configured to apply a prediction filtering to the spectral domain representation, in order to obtain a prediction filtered spectral domain representation) which is based on the filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware)) (e.g. apply a filtering in the frequency direction (e.g. a prediction filtering adapted to perform a noise shaping) to the spectral domain representation, in order to obtain a (prediction) filtered spectral domain representation using a filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware)) and / or using a prediction processing structure.

[0052] Furthermore, the encoder is configured to obtain an encoded representation on the basis of the filtered and / or prediction processed spectral domain representation (e.g. in a bitstream), if the encoder is in the noise shaping operating mode

[0053] Furthermore, the encoder is configured to (e.g. in the prediction gain operating mode) obtain (e.g. determine, e.g. to calculate) a precision reduced spectral domain representation (e.g. a precision reduced spectrum) of the signal based on the spectral domain representation, to obtain a filtered and / or prediction processed precision reduced spectral domain representation (e.g. a filtered version of the precision reduced spectral domain representation; e.g. a TNS filtered spectrum with reduced precision) on the basis of the precision reduced spectral domain representation using the filter structure, (e.g. using a filtering (e.g. a prediction filtering) in a frequency direction (e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering; e.g. using a filtering of the precision reduced spectral domain representation

[0054] FH240905PEP-2025315045. DOCX using a TNS filter; e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value; wherein, for example, the residual spectral domain representation is left out of consideration when obtaining the filtered precision reduced spectral domain representation) (wherein, for example, the encoder is configured to apply a prediction filter operating in the frequency direction to the precision reduced spectral domain representation (but not to the residual spectral domain representation), to obtain the filtered precision reduced spectral domain representation) (wherein, for example, the encoder is configured to filter the precision reduced spectral domain representation (but not the residual spectral domain representation) using a prediction filter operating in the frequency direction, to obtain the filtered precision reduced spectral domain representation; wherein, for example, the encoder is configured to apply a prediction filtering to the precision reduced spectral domain representation, in order to obtain a prediction filtered precision reduced spectral domain representation)) (e.g. apply a prediction filtering (e.g. a prediction filtering adapted to perform a noise shaping) to the precision reduced spectral domain representation in order to obtain a prediction filtered precision reduced spectral domain representation using the filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware)) and / or using the prediction processing structure.

[0055] Furthermore, the encoder is configured to obtain an encoded representation on the basis of the filtered and / or prediction processed precision reduced spectral domain representation (wherein, for example, the encoder is configured to encode the filtered precision reduced spectral domain representation, or a processed version of the precision reduced spectral domain representation, or a part of the precision reduced spectral domain representation), if the encoder is in the redundancy reduction operating mode.

[0056] It was recognized that a same filter and / or prediction processing structure may be used for distinct operating modes, in the form of the noise shaping operating mode and the redundancy reduction operating mode.

[0057] For example, depending on the input provided to said filter and / or prediction processing structure the different operating modes may be realized, e.g. instead of a more complicated or more complex use of different filtering and / or prediction processing schemes.

[0058] FH240905PEP-2025315045. DOCX Hence, for example, in the noise shaping operating mode, the full spectral domain representation of the signal may be filtered, so that decoder sided, quantization noise is shaped by a respective inverse filter.

[0059] However, by obtaining the precision reduced spectral domain representation of the signal (a transmission of which may require less bits, compared to the full spectral domain representation of the signal), the filtering / prediction processing unit may shift its functionality (when considered together with the respective inverse filtering / prediction processing in a respective decoder) from a noise shaping unit (as there may not be enough bitrate to transmit the full filtered spectral domain representation of the signal) to a redundancy reduction unit (e.g. wherein only signal energy is reduced and added again without loss of information on a decoder side).

[0060] Hence, a same filtering / prediction processing structure may be used for different operating modes, e.g. depending on the input provided to the structure.

[0061] Optionally, the filter / prediction processing structure may even use a same parametrization (e.g. filter / prediction processing parameters), irrespective of the current operating mode of the encoder.

[0062] However, optionally, depending on the operating mode, the filtering / prediction processing may include a rounding operation, e.g. to enable a lossless reconstruction of the coded signal.

[0063] With respect to the above-discussed embodiments, it is to be noted that, as an optional feature, the filtering and / or prediction processing may, for example, comprise or may, for example, be implemented as or may, for example, be one or more of the following: Inter-channel prediction, Joint stereo coding, Mid / side coding, Complex stereo prediction, Multi-channel coding tool; Long Term Prediction; Harmonic Prediction; Temporal Noise Shaping; Frequency Domain Prediction.

[0064] Hence, encoders (e.g. audio encoders) according to embodiments may be configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (e.g. in the form of a prediction processed precision reduced spectral domain representation) and / or the filtered and / or prediction processed spectral domain representation (e.g. in the form of a prediction processed spectral domain representation) using one or more of: Inter-channel prediction, Joint stereo coding, Mid / side coding, Complex stereo prediction, Multi-channel

[0065] FH240905PEP-2025315045. DOCX coding tool; Long Term Prediction; Harmonic Prediction; Temporal Noise Shaping; Frequency Domain Prediction.

[0066] Hence, as an example, a prediction processing in a frequency direction, and / or in a time direction and / or in a spatial direction and / or in a channel direction may be achieved, depending on the specific prediction or coding approach (or combinations thereof).

[0067] Hence, in other words, examples for a prediction processing or prediction, according to embodiments, e.g. in particular for audio coding, may comprise one or more of: Inter-channel prediction, Joint stereo coding, Mid / side coding, Complex stereo prediction, Multi-channel coding tool; Long Term Prediction; Harmonic Prediction; Temporal Noise Shaping; Frequency Domain Prediction.

[0068] Further embodiments according to the invention comprise a decoder for decoding a signal, wherein the decoder is configured to decode an encoded representation of a prediction information (e.g., input vector or input data to be processed as input, for example, by a DNN performing a prediction processing, e.g., control information for a DNN performing a prediction processing, e.g., a prediction processing residual, e.g., an excitation signal to be processed as input by a prediction processing structure, e.g., input vector or input data to be processed as input by a prediction processing structure, e.g., prediction coefficients), in order to obtain a decoded version of the prediction information, wherein the decoder is configured to decode an encoded representation of a residual information, in order to obtain a decoded version of the residual information, wherein the decoder is configured to obtain a prediction processed precision reduced spectral domain representation on the basis of the prediction information using a [e.g. decoder-sided] prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and wherein the decoder is configured to determine a combined spectrum based on the prediction processed precision reduced spectral domain representation and based on the decoded version of the residual information.

[0069] Further embodiments according to the invention comprise an encoder for encoding a signal, wherein the encoder is configured to obtain a spectral domain representation of the signal; wherein the encoder is configured to obtain a precision reduced spectral domain representation of the signal based on the spectral domain representation, wherein the encoder is configured to obtain a residual information based on the spectral domain representation; wherein the encoder is configured to obtain a prediction information (e.g., input vector or input data to be processed as input, for example, by a DNN performing a prediction processing, e.g.,

[0070] FH240905PEP-2025315045. DOCX control information for a DNN performing a prediction processing, e.g., a prediction processing residual, e.g., an excitation signal to be processed as input by a prediction processing structure, e.g., input vector or input data to be processed as input by a prediction processing structure, e.g., prediction coefficients,) on the basis of the precision reduced spectral domain representation using a (e.g. encoder-sided) prediction processing (e.g. an inverse prediction processing when considering a decoder-sided prediction processing, e.g. a prediction processing removing predictable signal portions; e.g., a prediction processing such that using a (e.g., decoder-sided) corresponding (e.g., forward instead of inverse) prediction processing the precision reduced spectral domain representation can be reconstructed) in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and wherein the encoder is configured to obtain an encoded representation on the basis of the prediction information and on the basis of the residual information.

[0071] It was recognized that the inventive determination (e.g. splitting or separation) of the precision reduced spectral domain representation and the residual information on the basis of the spectral domain representation of the signal may, for example, be applicable and, for example, advantageous with a method and / or apparatus (e.g., encoder-decoder concept), wherein, optionally, solely a prediction information is transmitted, based on which the prediction processed precision reduced spectral domain representation may be reconstructed (e.g. decoder-sided), for example without transmitting a spectral representation, such as a prediction residual.

[0072] This decoded prediction processed (e.g., decoder sided) (e.g. predicted) precision reduced spectral domain representation may be combined with the decoded residual information, which may represent a residual between the spectral domain representation and the precision reduced spectral domain representation (e.g. in contrast to a conventional prediction residual between the prediction processed precision reduced spectral domain representation and the precision reduced spectral domain representation), so as to obtain a decoded version of the spectral domain representation.

[0073] For example, based on the determination of the precision reduced spectral domain representation, e.g. performed as a MSB / LSB split based on an available bitrate, a prediction processing such as a deep neural network (DNN) may be applied, in order to obtain neural network control and / or input parameters which allow for a prediction processing, e.g. prediction (e.g. on a decoder-side), of the precision reduced spectral domain representation without a residual signal.

[0074] FH240905PEP-2025315045. DOCX Hence, instead of performing prediction processing based on the original (e.g., full precision) spectral domain representation which may result in some cases in prediction information requiring high bitrate, or which may require a complex and computationally demanding prediction processing structure (e.g., a large DNN), the precision reduced spectral domain representation and the residual information may be determined so that a prediction processing of the precision reduced spectral domain representation results in prediction information requiring a lower bitrate and / or prediction processing can be performed using a less computationally demanding prediction processing structure.

[0075] In addition, for example, similarly to what was outlined above, by transmitting the prediction information fully (e.g, losslessly) from the encoder to the decoderand, optionally, by performing the prediction processing independent of the residual information, a scalable (e.g., nearlossless) operation can be realized, where the effects of, for example, omitting the residual information, or, for example, a portion thereof, (e.g., dependent on an available bitrate) may, for example, be controlled precisely. The residual information may bypass the prediction processing step and may hence, although not necessarily transmitted fully (e.g. because of an available bitrate) not be subject to filter-based alterations of errors introduced due to the incomplete transmission.

[0076] In the following further embodiments according to the invention will be discussed. This following discussion is focused, for the sake of brevity, on a use of a filtering in frequency direction, e.g. in particular in the context of temporal noise shaping filter structures.

[0077] However, it is to be noted that the following aspects, functionalities and details, are to be understood in an according manner with respect to a use of a filtering in a time direction and / or in a spatial direction and / or in a channel direction and also, for example as an alternative or in combination with respect to a use of a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and also, for example, as an alternative or in combination with respect to a use of a prediction filtering or to a use of a predictive filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction.

[0078] However, according to at least some embodiments, a same processing structure may be considered a filtering structure (such as a noise shaping filter) and / or a prediction processing structure depending on the further processing of the respective signal down the coding chain, e.g. whether quantization errors are introduced and whether and how such errors are affected, e.g. shaped in a respective decoder (e.g. by respective inverse filtering / prediction processing).

[0079] FH240905PEP-2025315045. DOCX Hence, for obtaining a filtered and / or prediction processed version of a signal (e.g. as an example of the precision reduced spectral domain representation, or a decoded or encoded version thereof respectively) the following embodiments are also not limited to performing a filtering, as such filters may act as predictors (e.g. realizing a prediction and may hence be considered predictors or prediction processing structures) or as such filters may be prediction filters or predictive filters.

[0080] For example, depending on whether a processed version of a spectral domain representation is transmitted fully, e.g. losslessly, or in an incomplete manner, e.g. lossily, said processing may act as a filtering (e.g. shaping quantization noise by the filtering in the encoder and the inverse filtering in the decoder wherein, because of noise from the lossy transmission, said noise is shaped by the inverse filtering) or as a prediction (e.g. removing signal energy in the encoder and reintroducing said energy in a respective decoder).

[0081] However, in this regard, it is to be noted that according to some embodiments, a prediction may, for example, be considered (e.g. at least partly) as a separate and more general concept (e.g. compared to a filtering).

[0082] For example, the term ..filtering", e.g. in the audio signal processing context, may be understood at least in some cases, or may, for example comprise at least in some cases (or may, for example, at least in certain fields be commonly tied to the notion of) a linear filtering, e.g. a combination, e.g. a weighted combination, e.g. a weighted sum (i.e., a weighted sum / combination) of sample values that can mathematically be expressed as a convolution.

[0083] In line with this, according to at least some embodiments, the term ..prediction" or “prediction processing”, can be understood in a more general sense. A predictor or prediction processing structure according to some embodiments (hence, accordingly a prediction or a prediction processing) could also be, for example, a DNN (e.g. a deep neural network), or any other suitable algorithm for prediction processing (e.g. predicting, e.g., deriving, e.g. calculating, e.g., in some cases, approximating, e.g., estimating) a sample value (e.g., a spectral bin value) based on other information (e.g. such as other, e.g. previously coded sample values (e.g., spectral bin values).

[0084] Additionally, according to at least some embodiments, a prediction processing, e.g. a prediction, can optionally be based on (for example, other, e.g., neighboring, e.g., in a frequency direction and / or in a time direction and / or in a spatial direction, and / or in a channel

[0085] FH240905PEP-2025315045. DOCX direction) sample values (e.g., spectral bin values) alone (e.g. as may be the case with filtering), but could generally (and, for example, in contrast to filtering, e.g. at least some forms of filtering) also take other information, such as side-information, into account.

[0086] As a general remark, a linear filtering can, for example, be a special case of a prediction or prediction processing. For example, referring to Fig. 7, P(z) may be a linear filtering, the filter P(z) being a Predictor (hence “P” or rather P(z)). This may, for example, be a special case of prediction, but prediction is, for example according to some embodiments, more general and not limited to this special case.

[0087] As a general remark, also conversely, not all filters are suitable for performing a prediction processing.

[0088] Hence, at least some embodiments do generally, e.g. as an optional feature, comprise a prediction processing structure that is not necessarily a filter or even a linear filter.

[0089] With this in mind, the following embodiments are to be understood in an according manner with respect to a use of a prediction processing, e.g. as a generalization of or even as a different approach to a filtering, e.g. with decoder and encoder performing a prediction processing and an inverse prediction processing instead of or in addition to a filtering and an inverse filtering.

[0090] Furthermore, in the following, mainly filtering in a frequency direction is discussed. However, in the same manner, e.g. interchangeably, embodiments may comprise a filtering and / or prediction in a time direction, e.g. processing or using information of one or more previous frames.

[0091] Accordingly, a prediction processing and / or filtering may be performed across different channels (e.g. for multi-channel applications, e.g. performing a filtering and / or prediction over channels, e.g., predicting the SL channel from FL and SR, for example) and / or with respect to signals or signal portions associated to different spatial domains.

[0092] Embodiments according to the invention comprise a decoder for decoding a signal, wherein the decoder is configured to decode an encoded representation of a precision reduced spectral domain representation of a signal (e.g. a filtered version of a precision reduced spectral domain representation, e.g. as provided in the encoder; e.g. a TNS filtered spectrum with reduced precision; e.g. a more significant spectral value information; e.g. from a bitstream), in order to

[0093] FH240905PEP-2025315045. DOCX obtain a decoded version of the precision reduced spectral domain representation, wherein the decoder is configured to decode an encoded representation of a residual information (e.g. a residual spectral domain representation or a portion thereof, e.g. a residual between a spectral domain representation and the precision reduced spectral domain representation; which is, for example, a residual of the precision reduced spectral domain representation; which may, for example, represent a difference between the spectral domain representation and the precision reduced spectral domain representation; e.g. a residual of reduced spectrum; e.g. a residual between the spectral domain representation and the precision reduced spectral domain representation; e.g. a less significant spectral value information, e.g. corresponding to the more significant spectral value information; e.g. from a bitstream), in order to obtain a decoded version of the residual information.

[0094] Furthermore, the decoder is configured to obtain a filtered precision reduced spectral domain representation (e.g. the precision reduced spectral domain representation) on the basis of the precision reduced spectral domain representation using a filtering (e.g. an inverse filtering; e.g. an inverse prediction filtering; e.g. a prediction filtering) in a frequency direction (e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering, e.g. using a frequency domain inverse prediction filtering; e.g. using a filtering of the precision reduced spectral domain representation using a TNS filter, e.g. using an inverse TNS filter; e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value)(wherein, for example, the encoder is configured to apply a prediction filter, e.g. an inverse prediction filter, operating in the frequency direction to the filtered precision reduced spectral domain representation, to obtain the precision reduced spectral domain representation)(wherein, for example, the encoder is configured to filter the filtered precision reduced spectral domain representation using a prediction filter operating in the frequency direction, to obtain the precision reduced spectral domain representation) (wherein, for example, the encoder is configured to apply a prediction filtering, e.g. an inverse prediction filtering, to the filtered precision reduced spectral domain representation, in order to obtain a precision reduced spectral domain representation)(e.g. apply an inverse prediction filtering to the decoded version of the prediction filtered precision reduced spectral domain representation, in order to obtain the precision reduced spectral domain representation).

[0095] Furthermore, the decoder is configured to determine a combined spectrum based on the filtered precision reduced spectral domain representation and the decoded version of the residual information.

[0096] FH240905PEP-2025315045. DOCX According to an embodiment of the invention, the decoded version of the residual information represents a less significant (e.g. low-order) portion of a number representation of a spectral domain representation of the audio signal, wherein the filtered precision reduced spectral domain representation represents a more significant (e.g. high-order) portion of the number representation of the spectral domain representation of the audio signal, and the decoder is configured to combine the filtered precision reduced spectral domain representation and the decoded version of the residual information, in order to obtain the number representation (e.g. the envisioned number representation) of the spectral domain representation of the audio signal.

[0097] According to an embodiment of the invention, the decoder is configured to obtain (e.g. to derive or to determine) a first number representation (for the more significant portion of the number representation of the spectral domain representation of the audio signal), which represents the filtered precision reduced spectral domain representation, wherein the decoder is configured to obtain (e.g. to derive or to determine) a second number representation (for the less significant portion of the number representation of the spectral domain representation of the audio signal), which represents the decoded version of the residual information, and wherein the decoder is configured to combine (e.g. concatenate; e.g. combine in a non-overlapping manner; e.g. combine in such a manner that the first number representation defines a first portion of the number representation of the spectral domain representation of the audio signal and such that the second number representation defines a second portion of the number representation of the spectral domain representation of the audio signal, wherein the second portion is different from the first portion) the first number representation and the second number representation, in order to obtain the number representation (e.g. the envisioned number representation) of the spectral domain representation of the audio signal.

[0098] According to an embodiment of the invention, the decoder is configured to concatenate (e.g. in the sense of a bit-wise combination) the first number representation and the second number representation, in order to obtain the number representation of the spectral domain representation of the audio signal (e.g. such that the first number representation defines a plurality of bit positions of the number representation of the spectral domain representation of the audio signal, and such that the second number representation defines a plurality of other bit positions of the number representation of the spectral domain representation of the audio signal).

[0099] FH240905PEP-2025315045. DOCX According to an embodiment of the invention, the decoded version of the residual information represents one or more least significant bits, LSBs, of the number representation of the spectral domain representation of the audio signal, wherein the filtered precision reduced spectral domain representation represents one or more most significant bits, MSBs, of the number representation of the spectral domain representation of the audio signal, and wherein the decoder is configured to combine the one or more LSBs and the one or more MSBs, in order to obtain the number representation of the spectral domain representation of the audio signal.

[0100] According to an embodiment of the invention, the decoder is configured to determine the decoded version of the residual information with variable accuracy (of the decoded version of the residual information), depending on an amount (e.g. a number of encoded bits) of residual information included in the encoded representation of the residual information (wherein, for example, the decoder is configured to handle cases in which a number of least significant bits included in the encoded representation of the residual information is smaller than expected in view of a significance (e.g. bit weight) of the more significant (e.g. high-order) portion of a number representation of the spectral domain representation) (wherein the decoder is configured to handle cases in which the more significant (e.g. high-order) portion of a number representation of the spectral domain representation is fully encoded and in which one or more associated least significant bits are not encoded within the encoded representation of the residual information, e.g. in order to save bit rate) (so that, as an example, if the number representation of the spectral domain representation has 5 digits with MSBs defining digits 3 to 5 and LSBs defining digits 1 to 2, the decoder may adapt the determination of the number representation of the spectral domain representation, if instead of the 2 expected LSBs (e.g. expected because of the 5 total digits only the 3 MSBs were received), less than 2 LSBs or even no LSBs were received, e.g. so as to fill up missing LSBs with default values to approximate the number representation of the spectral domain representation with the received MSBs and the default values).

[0101] According to an embodiment of the invention, the decoded version of the residual information comprises a residual spectral domain representation, representing a difference between the spectral domain representation and the filtered precision reduced spectral domain representation.

[0102] According to an embodiment of the invention, the decoder is configured to obtain the filtered precision reduced spectral domain representation using a prediction filtering in a frequency direction.

[0103] FH240905PEP-2025315045. DOCX According to an embodiment of the invention, the decoder is configured to obtain the filtered precision reduced spectral domain representation using a temporal noise shaping filter structure, in order to effect a prediction filtering in a frequency direction of the precision reduced spectral domain representation.

[0104] According to an embodiment of the invention, the decoder is configured to obtain the filtered precision reduced spectral domain representation using a filter structure, wherein the decoder is configured to decode (or for example to receive, or for example to obtain) a set of filter parameters (e.g. LPC parameters) (e.g. parameters describing a weighting of one or more previously processed spectral values / spectral bin values in a prediction) for use with the filter structure, (wherein the filtering structure may be configured to effect a temporal noise shaping and / or achieve a prediction gain in dependence on the set of filter parameters and / or in dependence on a mode flag, e.g. a mode flag, indicating whether a rounding operation of the filter should be activated or deactivated) (e.g. such that the filtering structure effects a temporal noise shaping or achieves a prediction gain in dependence on the precision reduced spectral domain representation, e.g. in dependence on whether the precision reduced spectral domain representation comprises a noise contribution or not).

[0105] According to an embodiment of the invention, the decoder is configured to decode filter parameters, and wherein the decoder is configured to obtain the filtered precision reduced spectral domain representation using a filter structure, wherein the decoder is configured to adapt the filtering performed using the filter structure based on the filter parameters, in order to adapt a weighting of a prediction of the filtering, and / or to a adapt a prediction order of the filtering and / or in order to reconfigure the filter structure.

[0106] According to an embodiment of the invention, the decoded version of the precision reduced spectral domain representation comprises integer-valued spectral bin values for a plurality of spectral bins, and wherein the decoder is configured to obtain the filtered precision reduced spectral domain representation (e.g. the precision reduced spectral domain representation) in the form of a integer valued filtered precision reduced spectral domain representation (e.g. using a rounding operation; e.g. using a filtering comprising a rounding operation) (wherein, optionally, the decoded version of the precision reduced spectral domain representation comprises integer-valued spectral bin values for a plurality of spectral bins) (wherein, for example, the filtered precision reduced spectral domain representation (also) comprises integer-valued spectral bin values for a plurality of spectral bins).

[0107] FH240905PEP-2025315045. DOCX According to an embodiment of the invention, the decoded version of the precision reduced spectral domain representation comprises integer-valued spectral bin values for a plurality of spectral bins, wherein the decoder is configured to map integer values of the decoded version of the precision reduced spectral domain representation onto integer values of the filtered precision reduced spectral domain representation using a rounding of predicted values (which are combined with, e.g. added to or subtracted from, respective values of the decoded version of the precision reduced spectral domain representation, in order to obtain respective values of the filtered precision reduced spectral domain representation) or using a rounding of result spectral bin values, resulting from combinations of respective currently considered spectral bin values and respective predicted values).

[0108] According to an embodiment of the invention, the decoded version of the precision reduced spectral domain representation of a signal is an energy-reduced version of a precision reduced spectral domain representation of the signal, and wherein the decoder is configured to losslessly reconstruct the precision reduced spectral domain representation of the signal using the filtering (e.g. in the form of the filtered precision reduced spectral domain representation; e.g. in that the filtering adds a predicted spectral contribution, obtained using prediction filter parameters on the basis of the decoded version of the precision reduced spectral domain representation of the signal to the decoded version of the precision reduced spectral domain representation of the signal) (e.g. so that the filter effects “lossless reconstruction” of integer values (of more significant portion of integer values)).

[0109] According to an embodiment of the invention, the decoder is configured to obtain the filtered precision reduced spectral domain representation on the basis of the precision reduced spectral domain representation using a filtering which is independent of the residual information (e.g. so that the residual information, e.g. LSBs, do not influence said filtering, e.g. so that LSBs do not influence a state of a respective filter used for the filtering) (e.g. so that a potential error signal embodied by the residual information is not temporally noise shaped and is not processed by the filtering).

[0110] According to an embodiment of the invention, the decoder is configured to decode a filter mode information, and wherein the decoder is configured to adapt the filtering of the precision reduced spectral domain representation in order to obtain the filtered precision reduced spectral domain representation based on the filter mode information (e.g. in order to activate or deactivate a rounding functionality, e.g. in order to perform an inverse TNS filtering with or without rounding).

[0111] FH240905PEP-2025315045. DOCX According to an embodiment of the invention, the decoder is configured to decode an information describing the operation mode of the encoder used to encode the encoded representation of a precision reduced spectral domain representation (e.g. an information about the operating mode of the encoder; e.g. an information indicating the operating mode of the encoder; e.g. one or more bits, e.g. dedicated bits, indicating the operation mode of the encoder); and wherein the decoder is configured to adapt the filtering of the precision reduced spectral domain representation in order to obtain the filtered precision reduced spectral domain representation based on the information describing the operating mode of the encoder (e.g. in order to activate or deactivate a rounding functionality), and / or wherein the decoder is configured to selectively determine the combined spectrum based on the filtered precision reduced spectral domain representation and the decoded version of the residual information or based on the filtered precision reduced spectral domain representation without considering a residual information (and e.g. hence to skip obtaining a decoded version thereof).

[0112] According to an embodiment of the invention, the decoder is configured use a default residual information instead of the decoded version of the residual information, if the encoded representation of a residual information is absent in a respective bitstream.

[0113] According to an embodiment of the invention, the decoder is configured to selectively enable or disable usage of a rounding functionality in the filtering in the frequency direction (e.g. in order to switch between a lossless or near lossless spectrum reconstruction (e.g. of the filtered precision reduced spectral domain representation) and a lossy spectrum reconstruction (e.g. of the spectral domain representation)).

[0114] According to an embodiment of the invention, the filtering in the frequency direction is configured to perform a lossless reconstruction of the filtered precision reduced spectral domain representation (e.g. such that the filtered precision reduced spectral domain representation obtained in the decoder is identical to a precision reduced spectral domain representation to which a prediction filtering is applied at the side of an encoder; e.g. in that the filtering in the frequency direction performed in the decoder is configured to reverse (e.g. precisely reverse, losslessly reverse) a filtering in the frequency direction performed at the side of an encoder).

[0115] According to an embodiment of the invention, the filtering in the frequency direction is a TNS (temporal noise shaping) filter with lossless reconstruction.

[0116] FH240905PEP-2025315045. DOCX According to an embodiment of the invention, the decoder is configured to receive an information (e.g. a side information; e.g. a side information included in a bit stream comprising the encoded representation of a precision reduced spectral domain representation of a signal and the encoded representation of the residual information) describing a splitting between the representation of the precision reduced spectral domain representation of the signal and the representation of the residual information (e.g. a number of bits of the combined spectrum represented by the encoded representation of the precision reduced spectral domain representation and / or a number of bits of the combined spectrum represented by the encoded representation of the residual information).

[0117] According to an embodiment of the invention, the decoder is configured to receive an information (e.g. a side information; e.g. a side information included in a bit stream comprising the encoded representation of a precision reduced spectral domain representation of a signal and the encoded representation of the residual information) describing a whether a rounding should be applied in the filtering in the frequency direction (e.g. in order to round a predicted value to an integer value).

[0118] According to an embodiment of the invention, the decoder is configured to switch (at least) between a lossless mode, a near lossless mode and a lossy mode (e.g. in dependence on an operation mode information; e.g. included in the bitstream; e.g. an information describing the operation mode of the encoder; e.g. a flag); wherein the decoder is configured to decode and to losslessly (e.g. using a rounding) filter a (e.g. full, e.g. complete, e.g. losslessly encoded) spectral domain representation (e.g. a complete representation of spectral domain values), if the decoder is in the lossless mode, wherein the decoder is configured to decode and to losslessly (e.g. using a rounding) filter the precision reduced spectral domain representation (and to supplement then filtered precision reduced spectral domain representation using the residual spectral domain representation as far as an encoded residual spectral domain representation is received by the decoder), if the encoder is in the near lossless mode, and wherein the decoder is configured to decode and to lossily (e.g. without using a rounding) filter a (e.g. lossily encoded) spectral domain representation (e.g. the precision reduced spectral domain representation), if the encoder is in the lossy mode.

[0119] Embodiments according to the invention comprise an encoder for encoding a signal (e.g. an audio signal), wherein the encoder is configured to obtain (e.g. derive, e.g. receive) a spectral domain representation (e.g. an initial spectral domain representation; e.g. an original precision spectral domain representation; e.g. a spectral domain representation in a full integer value representation; e.g. an MDCT spectrum) of the signal; wherein the encoder is configured to

[0120] FH240905PEP-2025315045. DOCX obtain (e.g. determine, e.g. to calculate) a precision reduced spectral domain representation (e.g. a precision reduced spectrum) of the signal based on the spectral domain representation.

[0121] Furthermore, the encoder is configured to obtain (e.g. determine, e.g. to calculate, e.g. to approximate, e.g. to set, e.g. to estimate) a residual information (e.g. a residual spectral domain representation or a portion thereof, e.g. a residual between the spectral domain representation and the precision reduced spectral domain representation; which is, for example, a residual of the precision reduced spectral domain representation; which may, for example, represent a difference between the spectral domain representation and the precision reduced spectral domain representation; e.g. a residual of reduced spectrum) based on (e.g. on the basis of; e.g. in dependence on) the spectral domain representation; (wherein the encoder may, for example, be configured to split up the spectral domain representation of the signal into the precision reduced spectral domain representation and a residual of the precision reduced spectral domain representation, in order to obtain the precision reduced spectral domain representation and the residual information) wherein the encoder is configured to obtain a filtered precision reduced spectral domain representation (e.g. a filtered version of the precision reduced spectral domain representation; e.g. a TNS filtered spectrum with reduced precision) on the basis of the precision reduced spectral domain representation using a filtering (e.g. a prediction filtering) in a frequency direction (e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering; e.g. using a filtering of the precision reduced spectral domain representation using a TNS filter; e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value) (wherein, for example, the residual spectral domain representation is left out of consideration when obtaining the filtered precision reduced spectral domain representation) (wherein, for example, the encoder is configured to apply a prediction filter operating in the frequency direction to the precision reduced spectral domain representation (but not to the residual spectral domain representation), to obtain the filtered precision reduced spectral domain representation).

[0122] Furthermore, for example, the encoder is configured to filter the precision reduced spectral domain representation (but not the residual spectral domain representation) using a prediction filter operating in the frequency direction, to obtain the filtered precision reduced spectral domain representation; wherein, for example, the encoder is configured to apply a prediction filtering to the precision reduced spectral domain representation, in order to obtain a prediction filtered precision reduced spectral domain representation), wherein the encoder is configured

[0123] FH240905PEP-2025315045. DOCX to obtain an encoded representation on the basis of the filtered precision reduced spectral domain representation and the residual information (wherein, for example, the encoder is configured to encode the filtered precision reduced spectral domain representation, or a precessed (e.g. processed) version of the precision reduced spectral domain representation, or a part of the precision reduced spectral domain representation, and wherein, for example, the encoder is configured to encode the residual information, or a processed version of the residual information, or a part of the residual information (e.g. as an index to a table, e.g. a look-up table, comprising approximations or default values for residuals; e.g. in a bitstream)).

[0124] According to an embodiment of the invention, the spectral domain representation is an integer spectral domain representation of the signal (e.g. as an Integer MDCT spectrum or a portion thereof), wherein the encoder is configured to obtain (e.g. determine, e.g. to calculate) the precision reduced spectral domain representation (e.g. a precision reduced spectrum) of the signal in the form of an integer precision reduced spectral domain representation based on the integer spectral domain representation, and wherein the encoder is configured to obtain the filtered precision reduced spectral domain representation (e.g. a filtered version of the precision reduced spectral domain representation; e.g. a TNS filtered spectrum with reduced precision) as a filtered integer precision reduced spectral domain representation on the basis of the integer precision reduced spectral domain representation using a filtering (e.g. a prediction filtering) in a frequency direction.

[0125] According to an embodiment of the invention, the encoder is configured to obtain the filtered integer precision reduced spectral domain representation using a rounding operation (e.g. using a filtering comprising a rounding operation).

[0126] According to an embodiment of the invention, the encoder is configured to obtain (e.g. determine, e.g. to calculate) the precision reduced spectral domain representation (e.g. a precision reduced spectrum) in the form of a more significant portion of a number representation of the spectral domain representation of the audio signal.

[0127] According to an embodiment of the invention, the encoder is configured to obtain (e.g. determine, e.g. to calculate) the precision reduced spectral domain representation (e.g. a precision reduced spectrum) in the form of a first number representation (e.g. for the more significant portion of the number representation of the spectral domain representation of the audio signal).

[0128] FH240905PEP-2025315045. DOCX According to an embodiment of the invention, the encoder is configured to obtain (e.g. determine, e.g. to calculate) the precision reduced spectral domain representation (e.g. a precision reduced spectrum) in the form of most significant bits, MSBs, of the number representation of the spectral domain representation of the audio signal.

[0129] According to an embodiment of the invention, the encoder is configured to obtain (e.g. determine, e.g. to calculate) the residual information in the form of a residual spectral domain representation or a portion thereof based on the spectral domain representation.

[0130] According to an embodiment of the invention, the encoder is configured to obtain (e.g. determine, e.g. to calculate) the residual information in the form of a less significant (e.g. low- order) portion of a number representation of a spectral domain representation.

[0131] According to an embodiment of the invention, the encoder is configured to obtain (e.g. determine, e.g. to calculate) the residual information in the form of a second number representation (e.g. for the less significant portion of the number representation of the spectral domain representation of the audio signal) based on the spectral domain representation.

[0132] According to an embodiment of the invention, the encoder is configured to obtain (e.g. determine, e.g. to calculate) the residual information in the form of one or more least significant bits, LSBs, of the number representation of the spectral domain representation of the audio signal,

[0133] According to an embodiment of the invention, the encoder is configured to adapt the filtering in frequency direction based on the available bitrate.

[0134] According to an embodiment of the invention, the encoder is configured to selectively (e.g. switchably) use a filtering with rounding or a filtering without rounding, in order to obtain the filtered precision reduced spectral domain representation (e.g. depending on the available bitrate and / or depending on the signal).

[0135] According to an embodiment of the invention, the encoder is configured to use the filtering with rounding if an available bitrate is sufficient to losslessly encode the filtered precision reduced spectral domain representation obtainable by filtering with rounding, and wherein the encoder is configured to use the filtering without rounding if the available bitrate is insufficient to losslessly encode the filtered precision reduced spectral domain representation obtainable by filtering with rounding.

[0136] FH240905PEP-2025315045. DOCX According to an embodiment of the invention, the encoder is configured to obtain the filtered precision reduced spectral domain representation (e.g. a filtered version of the precision reduced spectral domain representation; e.g. a TNS filtered spectrum with reduced precision) on the basis of the precision reduced spectral domain representation using a temporal noise shaping filter structure, in order to effect a prediction filtering in a frequency direction of the precision reduced spectral domain representation.

[0137] According to an embodiment of the invention, the encoder is configured to obtain the filtered precision reduced spectral domain representation using a filter structure, wherein the encoder is configured to obtain (or for example to receive, or for example to obtain) a set of filter parameters (e.g. LPC parameters; e.g. parameters describing a weighting of one or more previously processed spectral values / spectral bin values in a prediction) for use with the filter structure, (wherein the filter structure may be configured to effect a temporal noise shaping and / or achieve a prediction gain in dependence on the set of filter parameters; wherein the encoder may be configured to encode a mode flag (e.g. indicating a temporal noise shaping mode, e.g. indicating a redundancy reduction operating mode, e.g. indicating respective lossy, lossless or nearly lossless versions thereof), e.g. a mode flag indicating whether a rounding operation of the filter should be activated or deactivated) (e.g. such that the filter structure effects a temporal noise shaping or achieves a prediction gain in dependence on the precision reduced spectral domain representation, e.g. so that the temporal noise shaping or the prediction gain are achieved for a filtered precision reduced spectral domain representation in a respective decoder, after applying a respective inverse filtering, e.g. in dependence on whether the precision reduced spectral domain representation on the decoder-side comprises a noise contribution or not).

[0138] According to an embodiment of the invention, the encoder is configured to use (e.g. as the filtering in frequency direction) a temporal noise shaping filtering with rounding, if an available bitrate is sufficient to losslessly encode the filtered precision reduced spectral domain representation obtainable by the temporal noise shaping filtering with rounding, and wherein the encoder is configured to use the temporal noise shaping filtering without rounding (e.g. as the filtering in frequency direction), if the available bitrate is insufficient to losslessly encode the filtered precision reduced spectral domain representation obtainable by the temporal noise shaping filtering with rounding.

[0139] According to an embodiment of the invention, the encoder is configured to encode (e.g. in the bitstream) a filter mode information indicating whether a filtering with rounding or a filtering

[0140] FH240905PEP-2025315045. DOCX without rounding was performed in order to obtain the filtered precision reduced spectral domain representation (and indicating whether a filtering with rounding or a filtering without rounding should be performed at the side of a decoder).

[0141] According to an embodiment of the invention, the encoder is configured to perform the filtering to obtain the filtered precision reduced spectral domain representation independent from the residual information (e.g. so that a prediction filtered version of MSBs (representing a more significant portion of the spectral domain representation) representing the filtered precision reduced spectral domain representation is obtained independently of LSBs (representing a less significant portion of the spectral domain representation) representing the residual information).

[0142] According to an embodiment of the invention, the encoder is configured to switch between (at least) different operating modes in dependence on an available bitrate.

[0143] According to an embodiment of the invention, the encoder is configured to switch (at least) between a lossless mode, a near lossless mode and a lossy mode in dependence on an available bitrate; wherein the encoder is configured to obtain and to encode a filtered version (e.g. a losslessly filtered version) of the spectral domain representation, if the encoder is in the lossless mode.

[0144] Furthermore, the encoder is configured to encode the filtered precision reduced spectral domain representation or at least a part thereof, (e.g. the losslessly filtered, precision reduced spectral domain representation or a part thereof; and, for example, to encode at least a part of the residual information; and, for example, to decide how much residual information is encoded in dependence on the available bitrate), if the encoder is in the near lossless mode, and wherein the encoder is configured to obtain and to encode a filtered version (e.g. a lossily filtered version) of the spectral domain representation or at least a part thereof, if the encoder is in the lossy mode.

[0145] According to an embodiment of the invention, the encoder is configured to switch (at least) between a lossless mode, a near lossless mode and a lossy mode in dependence on an available bitrate; wherein the encoder is configured to perform a filtering in frequency direction with rounding of the spectral domain representation in order to obtain a filtered version of the (e.g. full) spectral domain representation in the form of an integer representation and to encode the filtered version of the spectral domain representation (and to omit an encoding of a separate residual spectral domain representation), if the encoder is in the lossless mode.

[0146] FH240905PEP-2025315045. DOCX Furthermore, the encoder is configured to perform a filtering in frequency direction with rounding of the precision reduced spectral domain representation in order to obtain the filtered precision reduced spectral domain representation in the form of an integer representation and to encode the filtered precision reduced spectral domain representation, or at least a part thereof, (and, for example, to encode at least a part of the residual information; and, for example, to decide how much residual information is encoded in dependence on the available bitrate), if the encoder is in the near lossless mode, and wherein the encoder is configured to perform a filtering in frequency direction without rounding of the spectral domain representation in order to obtain a filtered version of the spectral domain representation and to encode the filtered version of the spectral domain representation, or at least a part thereof, if the encoder is in the lossy mode.

[0147] According to an embodiment of the invention, the encoder is configured to encode an information about the operating mode of the encoder.

[0148] According to an embodiment of the invention, the encoder is configured to determine the precision reduced spectral domain representation in dependence on an available bitrate and / or in dependence on the spectral domain representation (e.g. so as to determine the precision reduced spectral domain representation, and consequently the filtered precision reduced spectral domain representation, in order to losslessly transmit the filtered precision reduced spectral domain representation; e.g. to select a number of bits representing the precision reduced spectral domain representation (and possibly also the number of bits representing the residual information, which may be “complementary” to the number of bits representing the precision reduced spectral domain representation) in dependence on the available bitrate).

[0149] According to an embodiment of the invention, the encoder is configured to determine a bit representation (e.g. a binary representation; e.g. a m-ary representation with m>=2) of the filtered precision reduced spectral domain representation and to losslessly encode said bit representation.

[0150] According to an embodiment of the invention, the encoder is configured to determine a bit representation of the residual information and to selectively encode said bit representation (e.g. in full), a portion of said bit representation (e.g. a true subset of bits of said bit representation) or to selectively not encode said bit representation in dependence on an available bitrate (e.g. to selectively encode all bits of said bit representation, a true subset of the bits of said bit representation, or no bit of said bit representation).

[0151] FH240905PEP-2025315045. DOCX According to an embodiment of the invention, the encoder is configured to encode the filtered precision reduced spectral domain representation and the residual information separately in a bitstream (e.g. using different encoding functionalities; e.g. a scalable lossless to lossy coder and a residual coder).

[0152] According to an embodiment of the invention, the encoder comprises a dedicated functionality for encoding the residual information, wherein said functionality is configured to adapt (e.g. to scale, e.g. by truncating) a size of the residual information (e.g. corresponding to an accuracy of the residual information) or an amount of residual information that is encoded.

[0153] Embodiments according to the invention comprise an encoder, for encoding a signal (e.g. an audio signal), wherein the encoder is configured to switch between a noise shaping operating mode (e.g. a TNS operating mode; e.g. an operating mode in which the coefficients of the filter structure are determined such that the filter effects a temporal noise shaping, e.g. a temporal noise shaping in which a quantization noise of a decoded signal is reduced before a transient, or a temporal noise shaping in which a temporal evolution of the quantization noise of a decoded signal is adapted to (e.g. approximately follows) a signal envelope) (e.g. a lossy mode and / or a near lossless mode according to a Fallback strategy), and a redundancy reduction operating mode (e.g. a prediction gain operating mode; e.g. a lossless mode and / or a near lossless mode according to an expected case).

[0154] Furthermore, the encoder is configured to obtain (e.g. derive, e.g. receive) a spectral domain representation of the signal (e.g. an initial spectral domain representation; e.g. an original precision spectral domain representation; e.g. a spectral domain representation in a full integer value representation; e.g. an MDCT spectrum); wherein the encoder is configured to (e.g. in the TNS operating mode) obtain a filtered spectral domain representation (e.g. a filtered version of the spectral domain representation; e.g. a TNS filtered spectrum) on the basis of the spectral domain representation using a filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware) (e.g. using a filtering (e.g. a prediction filtering) in a frequency direction (e.g. a prediction filtering adapted to perform a noise shaping; e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering; e.g. using a filtering of the spectral domain representation using a TNS filter;e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value; wherein,

[0155] FH240905PEP-2025315045. DOCX for example, no residual spectral domain representation is determined or considered when obtaining the filtered spectral domain representation) (wherein, for example, the encoder is configured to apply a prediction filter operating in the frequency direction to the spectral domain representation, to obtain the filtered spectral domain representation) (wherein, for example, the encoder is configured to filter the spectral domain representation using a prediction filter operating in the frequency direction, to obtain the filtered spectral domain representation; wherein, for example, the encoder is configured to apply a prediction filtering to the spectral domain representation, in order to obtain a prediction filtered spectral domain representation) which is based on the filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware) (e.g. apply a filtering in the frequency direction (e.g. a prediction filtering adapted to perform a noise shaping) to the spectral domain representation, in order to obtain a (prediction) filtered spectral domain representation using a filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware)), and obtain an encoded representation on the basis of the filtered spectral domain representation (e.g. in a bitstream), if the encoder is in the noise shaping operating mode.

[0156] Furthermore, the encoder is configured to (e.g. in the prediction gain operating mode) obtain (e.g. determine, e.g. to calculate) a precision reduced spectral domain representation (e.g. a precision reduced spectrum) of the signal based on the spectral domain representation, to obtain a filtered precision reduced spectral domain representation (e.g. a filtered version of the precision reduced spectral domain representation; e.g. a TNS filtered spectrum with reduced precision) on the basis of the precision reduced spectral domain representation using the filter structure, (e.g. using a filtering (e.g. a prediction filtering) in a frequency direction (e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering; e.g. using a filtering of the precision reduced spectral domain representation using a TNS filter;e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value; wherein, for example, the residual spectral domain representation is left out of consideration when obtaining the filtered precision reduced spectral domain representation).

[0157] Furthermore, for example, the encoder is configured to apply a prediction filter operating in the frequency direction to the precision reduced spectral domain representation (but not to the residual spectral domain representation), to obtain the filtered precision reduced spectral

[0158] FH240905PEP-2025315045. DOCX domain representation) (wherein, for example, the encoder is configured to filter the precision reduced spectral domain representation (but not the residual spectral domain representation) using a prediction filter operating in the frequency direction, to obtain the filtered precision reduced spectral domain representation; wherein, for example, the encoder is configured to apply a prediction filtering to the precision reduced spectral domain representation, in order to obtain a prediction filtered precision reduced spectral domain representation), (e.g. apply a prediction filtering (e.g. a prediction filtering adapted to perform a noise shaping) to the precision reduced spectral domain representation in order to obtain a prediction filtered precision reduced spectral domain representation using the filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware) and to obtain an encoded representation on the basis of the filtered precision reduced spectral domain representation (wherein, for example, the encoder is configured to encode the filtered precision reduced spectral domain representation, or a precessed (e.g. processed) version of the precision reduced spectral domain representation, or a part of the precision reduced spectral domain representation), if the encoder is in the redundancy reduction operating mode.

[0159] According to an embodiment of the invention, the encoder is configured to obtain the filtered spectral domain representation in the noise shaping operating mode and the filtered precision reduced spectral domain representation in the redundancy reduction operating mode using the same filter structure and filter parameters, which are independent of the whether the encoder is in the noise shaping operating mode or in the redundancy reduction operating mode (e.g. using a same set of time-dependent, e.g. per-frame-provided, filter parameters, irrespective of whether the noise shaping operating mode or the redundancy reduction operating mode is used; e.g. using same filter parameters; e.g. so that a set of filter parameters provided via a bitstream is used indiscriminately to obtain the filtered spectral domain representation in the noise shaping operating mode or to obtain the filtered precision reduced spectral domain representation in the redundancy reduction operating mode).

[0160] According to an embodiment of the invention, the encoder is configured to obtain the filtered spectral domain representation in the noise shaping operating mode and the filtered precision reduced spectral domain representation in the redundancy reduction operating mode using the same filtering.

[0161] According to an embodiment of the invention, the encoder is configured to obtain (e.g. determine, e.g. to calculate, e.g. to approximate, e.g. to set, e.g. to estimate) a residual information (e.g. a residual spectral domain representation or a portion thereof, e.g. a residual

[0162] FH240905PEP-2025315045. DOCX between the spectral domain representation and the precision reduced spectral domain representation; which is, for example, a residual of the precision reduced spectral domain representation; which may, for example, represent a difference between the spectral domain representation and the precision reduced spectral domain representation; e.g. a residual of reduced spectrum) based on (e.g. on the basis of; e.g. in dependence on) the spectral domain representation and to obtain an encoded representation on the basis of the filtered precision reduced spectral domain representation and on the basis of the residual information (e.g. in the bitstream; e.g. as an index to a table, e.g. a look-up table, comprising approximations or default values for residuals), if the encoder is in the redundancy reduction operating mode.

[0163] According to an embodiment of the invention, the encoder is configured to switch between the operating modes in dependence on an available bitrate.

[0164] According to an embodiment of the invention, the encoder is configured to switch (at least) between a lossless mode, a near lossless mode and a lossy mode in dependence on an available bitrate; wherein the encoder is configured to obtain and to encode a filtered version (e.g. a losslessly filtered version) of the spectral domain representation, if the encoder is in the lossless mode.

[0165] Furthermore, the encoder is configured to encode the filtered precision reduced spectral domain representation or at least a part thereof (e.g. the losslessly filtered, precision reduced spectral domain representation or a part thereof; and, for example, to encode at least a part of the residual information; and, for example, to decide how much residual information is encoded in dependence on the available bitrate), if the encoder is in the near lossless mode, and wherein the encoder is configured to obtain and to encode a filtered version (e.g. a lossily filtered version) of the spectral domain representation or at least a part thereof, if the encoder is in the lossy mode.

[0166] According to an embodiment of the invention, the encoder is configured to switch at least between a lossy version of the noise shaping operating mode, a near lossless version of the noise shaping operating mode, a near lossless version of the redundancy reduction operating mode, and a lossless version of the redundancy reduction operating mode in dependence on an available bitrate.

[0167] Furthermore, the encoder is configured to perform a filtering in frequency direction with rounding (e.g. a lossless filtering) of the spectral domain representation in order to obtain a filtered version of the (e.g. full) spectral domain representation in the form of an integer

[0168] FH240905PEP-2025315045. DOCX representation and to encode the filtered version of the spectral domain representation (and to omit an encoding of a separate residual spectral domain representation), if the encoder is in the lossless version of the redundancy reduction operating mode.

[0169] Furthermore, the encoder is configured to perform the filtering in frequency direction with rounding (e.g. a lossless filtering) of the precision reduced spectral domain representation, in order to obtain the filtered precision reduced spectral domain representation in the form of an integer representation and to encode the filtered precision reduced spectral domain representation, or at least a part thereof (and, for example, to encode at least a part of the residual information; and, for example, to decide how much residual information is encoded in dependence on the available bitrate), if the encoder is in the near lossless version of the redundancy reduction operating mode.

[0170] Furthermore, the encoder is configured to perform the filtering in frequency direction with rounding (e.g. a lossless filtering) of the precision reduced spectral domain representation, in order to obtain the filtered precision reduced spectral domain representation in the form of an integer representation and to encode a portion (e.g. as a true subset) of the filtered precision reduced spectral domain representation (e.g. to lossily encode the filtered precision reduced spectral domain representation) but no residual information, if the encoder is in the near lossless version of the noise shaping operating mode.

[0171] Furthermore, the encoder is configured to perform a filtering in frequency direction without rounding (e.g. a lossy filtering) of the spectral domain representation, in order to obtain a filtered version of the spectral domain representation and to encode a portion (e.g. as a true subset) of the filtered version of the spectral domain representation, if the encoder is in the lossy version of the noise shaping operating mode.

[0172] According to an embodiment of the invention, the encoder is configured to encode an information about the operating mode (and optionally an information about the respective version of the operating mode) of the encoder.

[0173] Embodiments according to the invention comprise a method for decoding a signal, the method comprising: decoding an encoded representation of a precision reduced spectral domain representation of a signal (e.g. a filtered version of a precision reduced spectral domain representation, e.g. as provided in the encoder; e.g. a TNS filtered spectrum with reduced precision; e.g. a more significant spectral value information; e.g. from a bitstream), in order to obtain a decoded version of the precision reduced spectral domain representation, decoding

[0174] FH240905PEP-2025315045. DOCX an encoded representation of a residual information (e.g. a residual spectral domain representation or a portion thereof, e.g. a residual between a spectral domain representation and the precision reduced spectral domain representation; which is, for example, a residual of the precision reduced spectral domain representation; which may, for example, represent a difference between the spectral domain representation and the precision reduced spectral domain representation; e.g. a residual of reduced spectrum; e.g. a residual between the spectral domain representation and the precision reduced spectral domain representation; e.g. a less significant spectral value information, e.g. corresponding to the more significant spectral value information; e.g. from a bitstream), in order to obtain a decoded version of the residual information, obtaining a filtered precision reduced spectral domain representation (e.g. the precision reduced spectral domain representation) on the basis of the precision reduced spectral domain representation using a filtering (e.g. an inverse filtering; e.g. an inverse prediction filtering; e.g. a prediction filtering) in a frequency direction (e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering, e.g. using a frequency domain inverse prediction filtering; e.g. using a filtering of the precision reduced spectral domain representation using a TNS filter, e.g. using an inverse TNS filter; e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value).

[0175] Furthermore, for example, the method comprises applying a prediction filter, e.g. an inverse prediction filter, operating in the frequency direction to the filtered precision reduced spectral domain representation, to obtain the precision reduced spectral domain representation) (wherein, for example, the method comprises filtering the filtered precision reduced spectral domain representation using a prediction filter operating in the frequency direction, to obtain the precision reduced spectral domain representation; wherein, for example, the method comprises applying a prediction filtering, e.g. an inverse prediction filtering, to the filtered precision reduced spectral domain representation, in order to obtain a precision reduced spectral domain representation) (e.g. applying an inverse prediction filtering to the decoded version of the prediction filtered precision reduced spectral domain representation, in order to obtain the precision reduced spectral domain representation), and determining a combined spectrum based on the filtered precision reduced spectral domain representation and the decoded version of the residual information.

[0176] Embodiments according to the invention comprise a method for encoding a signal (e.g. an audio signal), the method comprising: obtaining (e.g. deriving, e.g. receiving) a spectral domain

[0177] FH240905PEP-2025315045. DOCX representation (e.g. an initial spectral domain representation; e.g. an original precision spectral domain representation; e.g. a spectral domain representation in a full integer value representation; e.g. an MDCT spectrum) of the signal; obtaining (e.g. determining, e.g. calculating) a precision reduced spectral domain representation (e.g. a precision reduced spectrum) of the signal based on the spectral domain representation, obtaining (e.g. determining, e.g. calculating, e.g. approximating, e.g. setting, e.g. estimating) a residual information (e.g. a residual spectral domain representation or a portion thereof, e.g. a residual between the spectral domain representation and the precision reduced spectral domain representation; which is, for example, a residual of the precision reduced spectral domain representation; which may, for example, represent a difference between the spectral domain representation and the precision reduced spectral domain representation; e.g. a residual of reduced spectrum) based on (e.g. on the basis of; e.g. in dependence on) the spectral domain representation.

[0178] Furthermore, the method may, for example, comprise splitting up the spectral domain representation of the signal into the precision reduced spectral domain representation and a residual of the precision reduced spectral domain representation, in order to obtain the precision reduced spectral domain representation and the residual information) obtaining a filtered precision reduced spectral domain representation (e.g. a filtered version of the precision reduced spectral domain representation; e.g. a TNS filtered spectrum with reduced precision) on the basis of the precision reduced spectral domain representation using a filtering (e.g. a prediction filtering) in a frequency direction (e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering; e.g. using a filtering of the precision reduced spectral domain representation using a TNS filter; e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value; wherein, for example, the residual spectral domain representation is left out of consideration when obtaining the filtered precision reduced spectral domain representation).

[0179] Furthermore, for example, the method comprises applying a prediction filter operating in the frequency direction to the precision reduced spectral domain representation (but not to the residual spectral domain representation), to obtain the filtered precision reduced spectral domain representation) (wherein, for example, the method comprises filtering the precision reduced spectral domain representation (but not the residual spectral domain representation) using a prediction filter operating in the frequency direction, to obtain the filtered precision reduced spectral domain representation; wherein, for example, the method comprises applying

[0180] FH240905PEP-2025315045. DOCX a prediction filtering to the precision reduced spectral domain representation, in order to obtain a prediction filtered precision reduced spectral domain representation), obtaining an encoded representation on the basis of the filtered precision reduced spectral domain representation and the residual information (wherein, for example, the method comprises encoding the filtered precision reduced spectral domain representation, or a precessed (e.g. processed) version of the precision reduced spectral domain representation, or a part of the precision reduced spectral domain representation, and wherein, for example, the method comprises encoding the residual information, or a processed version of the residual information, or a part of the residual information (e.g. as an index to a table, e.g. a look-up table, comprising approximations or default values for residuals; e.g. in a bitstream).

[0181] Embodiments according to the invention comprise a method for encoding a signal (e.g. an audio signal), the method comprising: switching between a noise shaping operating mode (e.g. a TNS operating mode; e.g. an operating mode in which the coefficients of the filter structure are determined such that the filter effects a temporal noise shaping, e.g. a temporal noise shaping in which a quantization noise of a decoded signal is reduced before a transient, or a temporal noise shaping in which a temporal evolution of the quantization noise of a decoded signal is adapted to (e.g. approximately follows) a signal envelope; e.g. a lossy mode or a near lossless mode according to a Fallback strategy), and a redundancy reduction operating mode (e.g. a prediction gain operating mode; e.g. a lossless mode or a near lossless mode according to an expected case); obtaining (e.g. deriving, e.g. receiving) a spectral domain representation of the signal (e.g. an initial spectral domain representation; e.g. an original precision spectral domain representation; e.g. a spectral domain representation in a full integer value representation; e.g. an MDCT spectrum).

[0182] Furthermore, the method further comprises, in the noise shaping operating mode, (e.g. in the TNS operating mode) obtaining a filtered spectral domain representation (e.g. a filtered version of the spectral domain representation; e.g. a TNS filtered spectrum) on the basis of the spectral domain representation using a filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware) (e.g. using a filtering (e.g. a prediction filtering) in a frequency direction (e.g. a prediction filtering adapted to perform a noise shaping; e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering; e.g. using a filtering of the spectral domain representation using a TNS filter; e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value; wherein,

[0183] FH240905PEP-2025315045. DOCX for example, no residual spectral domain representation is determined or considered when obtaining the filtered spectral domain representation).

[0184] Furthermore, for example, the method comprises applying a prediction filter operating in the frequency direction to the spectral domain representation, to obtain the filtered spectral domain representation) (wherein, for example, the method comprises filtering the spectral domain representation using a prediction filter operating in the frequency direction, to obtain the filtered spectral domain representation; wherein, for example, the method comprises applying a prediction filtering to the spectral domain representation, in order to obtain a prediction filtered spectral domain representation) which is based on the filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware) (e.g. applying a filtering in the frequency direction (e.g. a prediction filtering adapted to perform a noise shaping) to the spectral domain representation, in order to obtain a (prediction) filtered spectral domain representation using a filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware)), and obtaining an encoded representation on the basis of the filtered spectral domain representation (e.g. in a bitstream).

[0185] Furthermore, the method further comprises, in the redundancy reduction operating mode, (e.g. in the prediction gain operating mode) obtaining (e.g. determining, e.g. calculating) a precision reduced spectral domain representation (e.g. a precision reduced spectrum) of the signal based on the spectral domain representation, obtaining a filtered precision reduced spectral domain representation (e.g. a filtered version of the precision reduced spectral domain representation; e.g. a TNS filtered spectrum with reduced precision) on the basis of the precision reduced spectral domain representation using the filter structure, (e.g. using a filtering (e.g. a prediction filtering) in a frequency direction (e.g. over frequency; e.g. across frequency e.g. using a frequency domain prediction filtering; e.g. using a filtering of the precision reduced spectral domain representation using a TNS filter; e.g. using a filtering in which a prediction value obtained on the basis of one or more previous frequency bin values (e.g. previously processed frequency bin values) is subtracted from (or generally: combined with) a current (e.g. currently considered) frequency bin value, to obtain a processed current frequency bin value; wherein, for example, the residual spectral domain representation is left out of consideration when obtaining the filtered precision reduced spectral domain representation).

[0186] Furthermore, for example, the method comprises applying a prediction filter operating in the frequency direction to the precision reduced spectral domain representation (but not to the

[0187] FH240905PEP-2025315045. DOCX residual spectral domain representation), to obtain the filtered precision reduced spectral domain representation) (wherein, for example, the method comprises filtering the precision reduced spectral domain representation (but not the residual spectral domain representation) using a prediction filter operating in the frequency direction, to obtain the filtered precision reduced spectral domain representation; wherein, for example, the method comprises applying a prediction filtering to the precision reduced spectral domain representation, in order to obtain a prediction filtered precision reduced spectral domain representation), (e.g. applying a prediction filtering (e.g. a prediction filtering adapted to perform a noise shaping) to the precision reduced spectral domain representation in order to obtain a prediction filtered precision reduced spectral domain representation using the filter structure (e.g. a TNS filter structure; e.g. a TNS filter module implemented using software or using hardware, or using a combination of software and hardware)) and obtaining an encoded representation on the basis of the filtered precision reduced spectral domain representation (wherein, for example, the method comprises encoding the filtered precision reduced spectral domain representation, or a precessed (e.g. processed) version of the precision reduced spectral domain representation, or a part of the precision reduced spectral domain representation).

[0188] Embodiments according to the invention comprise a computer program for performing the method according to previous embodiments, when the computer program runs on a computer.

[0189] Embodiments according to the invention comprise a bitstream, comprising: an encoded representation of a precision reduced spectral domain representation of a signal, and an information (e.g. a dedicated bit) indicating whether a rounding should be applied in a filtering in a frequency direction that is to be to applied to a decoded representation of the precision reduced spectral domain representation in a decoder and optionally an encoded representation of a residual information, wherein the residual information represents a less significant (e.g. low-order) portion of a number representation of a spectral domain representation of the audio signal, wherein the precision reduced spectral domain representation represents a more significant (e.g. high-order) portion of the number representation of the spectral domain representation of the audio signal).

[0190] Embodiments according to the invention comprise a bitstream, comprising: an encoded representation of a precision reduced spectral domain representation of a signal, and an encoded representation of a residual information, and an information (e.g. a side information; e.g. a side information included in a bit stream comprising the encoded representation of a precision reduced spectral domain representation of a signal and the encoded representation of the residual information) describing a splitting between the representation of the precision

[0191] FH240905PEP-2025315045. DOCX reduced spectral domain representation of the signal and the representation of the residual information (e.g. a number of bits of the combined spectrum represented by the encoded representation of the precision reduced spectral domain representation and / or a number of bits of the combined spectrum represented by the encoded representation of the residual information).

[0192] Brief Description of the

[0193] The drawings are not necessarily to scale, emphasis instead generally being placed upon illustrating the principles of the invention. In the following description, various embodiments of the invention are described with reference to the following drawings, in which:

[0194] Fig. 1a shows a schematic view of a decoder according to embodiments of the invention;

[0195] Fig. 1b shows a schematic view of a decoder 100’ for decoding a signal using a prediction information according to an embodiment;

[0196] Fig. 2 shows a schematic view of a decoder with additional optional features according to embodiments of the invention;

[0197] Fig. 3a shows a schematic view of an encoder for encoding a signal according to embodiments of the invention;

[0198] Fig. 3b shows a schematic view of an encoder 300’ for encoding a signal using a prediction information according to an embodiment;

[0199] Fig. 4 shows a schematic view of an encoder with additional optional features, according to embodiments of the invention;

[0200] Fig. 5 shows a schematic view of another encoder according to embodiments of the invention;

[0201] Fig. 6 shows a schematic view of another encoder with additional, optional features, according to embodiments of the invention;

[0202] FH240905PEP-2025315045. DOCX Fig. 7 shows schematic views of different coding structures on which embodiments may be based;

[0203] Fig. 8 shows an example of how the MSE (Mean Squared Error) of spectral data is increased by an active TNS filter for some dedicated frames;

[0204] Fig. 9 shows a schematic view of a coding concept according to embodiments of the invention;

[0205] Fig. 10 shows a schematic view of an example of the data flow in all operation modes according to an embodiment; and

[0206] Fig. 11 shows a schematic scatter plot of segmental SNR values of frames where Temporal Noise Shaping (TNS) is active according to an embodiment.

[0207] Detailed Description of the Embodiments

[0208] Equal or equivalent elements or elements with equal or equivalent functionality are denoted in the following description by equal or equivalent reference numerals even if occurring in different figures.

[0209] In the following description, a plurality of details is set forth to provide a more thorough explanation of embodiments of the present invention. However, it will be apparent to those skilled in the art that embodiments of the present invention may be practiced without these specific details. In other instances, well-known structures and devices are shown in block diagram form rather than in detail in order to avoid obscuring embodiments of the present invention. In addition, features of the different embodiments described herein after may be combined with each other, unless specifically noted otherwise.

[0210] In the following, some inventive embodiments are explained in the context of a decoder and other inventive embodiments are explained in the context of an encoder. It is to be noted that features, functionalities and details that are explained in the context of a decoder may be implemented analogously in or added to or used with a corresponding encoder, individually or taken in combination. Vice versa, features, functionalities and details as disclosed for inventive encoders may be incorporated in corresponding decoders. Accordingly, it is to be noted that

[0211] FH240905PEP-2025315045. DOCX decoders and corresponding encoders (or vice versa) may be based on similar and / or equivalent inventive concepts and may hence comprise corresponding advantages.

[0212] Moreover, some inventive aspects are explained in the context of methods. It is to be noted that any of the features, functionalities and / or details as explained in the context of any of the inventive encoders and / or decoders may be incorporated in or may be used with or may be added to any of the inventive methods, individually or taken in combination. Furthermore, methods according the embodiments of the invention may be based on the same or similar or analogous considerations and / or ideas as corresponding encoders and / or decoders. Hence, these methods may comprise same or similar or analogous features and advantages.

[0213] According to the above explanations, some features, functionalities and details may be explained or disclosed in the context of embodiments according to a specific aspect, or an encoder rather than a decoder or vice versa, or according to a method, for the sake of brevity and conciseness. Hence, again, it is to be highlighted that any feature, functionality and / or detail of an embodiment may be incorporated or used with or added to any other embodiment according to the invention, individually or taken in combination.

[0214] The same applies accordingly to respective bitstreams and encoded representations.

[0215] Fig. 1a shows a schematic view of a decoder according to embodiments of the invention. Fig. 1a shows decoder 100, comprising decoding units 110 and 120, a filtering / prediction processing unit 130, which may comprise (e.g. which may be) a filtering unit (e.g., a prediction filtering unit, e.g. a predictive filtering unit), and / or a prediction processing unit, e.g. a prediction unit, e.g. a predictor unit and / or a filtering unit and a prediction processing unit, and a combining unit 140.

[0216] Decoder 100 is configured to decode a signal. Therefore, the decoder 100 is configured to obtain (e.g. to receive) an encoded representation 101 of a precision reduced spectral domain representation of a signal (e.g. of the signal to be decoded). For example, using the decoding unit 110, a decoded version 111 of the precision reduced spectral domain representation may be obtained.

[0217] Furthermore, the decoder 100 is configured to obtain (e.g. to receive) an encoded representation 102 of a residual information. For example, using the decoding unit 120 (e.g. a distinct residual decoder) a decoded version 121 of the residual information may be obtained. The decoded version 121 of the residual information may, for example, be or may, for example,

[0218] FH240905PEP-2025315045. DOCX comprise a residual spectral domain representation, representing a difference between the spectral domain representation and a filtered and / or prediction processed (e.g. predicted) precision reduced spectral domain representation 131.

[0219] Optionally, a decoder 100 according to embodiments may comprise different decoding units 110, 120, which may operate on different decoding mechanisms. Alternatively, the decoding units 110, 120, may operate on a same or similar approach or may even be implemented as a single decoding unit providing the two output signals 111 , 121.

[0220] Furthermore, for example using the filtering / prediction processing unit 130, the decoder is configured to obtain a filtered and / or prediction processed precision reduced spectral domain representation 131 on the basis of the precision reduced spectral domain representation (e.g. using the decoded version thereof 111).

[0221] Therefore, the decoder may be configured to filter the precision reduced spectral domain representation, e.g. 111 , in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction to determine the filtered and / or prediction processed precision reduced spectral domain representation 131. For example, the unit 130, may comprise a temporal noise shaping filter structure, configured to effect a prediction filtering in a frequency direction of the precision reduced spectral domain representation 111.

[0222] As an example (e.g. as an alternative), the decoder 100 may be configured to alternatively and / or in addition to the filtering, perform a prediction processing (e.g. a prediction step, e.g. a prediction, e.g. a reconstruction of previously removed redundant energy or information) on the basis of the precision reduced spectral domain representation, e.g. 111 , in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction to determine the filtered and / or prediction processed precision reduced spectral domain representation 131.

[0223] As an example, the decoder may apply a prediction filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction to determine the filtered and / or prediction processed precision reduced spectral domain representation 131.

[0224] For example, the filtering and / or prediction processing performed by the decoder, e.g. using unit 130, may be independent of the residual information. For example, prediction parameters or prediction coefficients and / or filtering parameters or filtering coefficients and / or a prediction state and / or a filter state may be independent of the residual information. Vice versa, the

[0225] FH240905PEP-2025315045. DOCX residual information may hence be unaffected by the filtering and / or prediction processing and hence in particular be unaffected by any form of noise shaping based on the filtering and / or prediction.

[0226] Furthermore, for example, using the combining unit 140, the decoder is configured to determine a combined spectrum 141 based on the filtered and / or prediction processed precision reduced spectral domain representation 131 and the decoded version of the residual information 121.

[0227] One important aspect may be that the precision reduced spectral representation is a precision reduced version of a spectral representation. Because it has reduced precision (e.g. based on a respective splitting of the full spectral representation), it can be transmitted losslessly with fewer bits than the full (i.e. full precision) representation. Another important aspect may be that the precision reduced representation may, for example, represent the more significant portion of the signal (e.g., MSBs, e.g., higher valued digits in a number representation, e.g., a representation obtained by dividing the original representation by a factor greater than 1 and truncating the result to a fixed number of digits, e.g., truncating it to an integer).

[0228] The residual information on the other hand may, for example, be or may, for example, represent the less significant signal portion.

[0229] Thus, the filtering and / or prediction processing may operate on the more significant signal portion. The more significant signal portion may still exhibit the trends, redundancy, patterns, predictability that the filtering and / or prediction processing concept exploits, whereas, for example, the residual information may represent the quasi-random fluctuation around the main trend and may therefore not, or to a lesser extent, exhibit exploitable trends, redundancy, patterns, predictability.)

[0230] As a side note, the precision reduced spectral domain representation may hence be considered a version of an original, e.g. underlying spectral domain representation, with reduced dynamic range. Hence, the decoder 100 may be provided with a split version of the spectral domain representation (e.g. of an approximation thereof, depending on quantization errors introduced in the residual information), in the form of the precision reduced spectral domain representation and the residual information.

[0231] The reduced precision (e.g. the reduced dynamic range) of the precision reduced spectral domain representation may enable avoiding an omission of (e.g. a need to omit) information

[0232] FH240905PEP-2025315045. DOCX (e.g. because of limited bitrate) from and thus an introduction of quantization noise in this portion of the original signal (e.g. this portion of the full spectral domain representation), so as to avoid an undesired influence (e.g. for cases wherein the filtering / prediction processing parameters are within certain constraints, e.g. having high gains) of the filtering / prediction processing unit of such potential quantization noise. Furthermore, this approach may allow exploiting the full potential of the filtering / prediction processing unit 130 with respect to a prediction gain, without the significant influence of such a prediction gain on potential quantization noise.

[0233] Hence, in the decoder 100, a very well-controlled processing of the spectral domain representation may be enabled by a splitting of the spectral domain representation to guarantee a well-known allocation of potential quantization noise, namely in the (e.g. unfiltered and not prediction processed, e.g. unpredicted) residual portion.

[0234] Hence, the signal coding chain shown on the top of Fig. 1a, e.g. resulting in the filtered and / or prediction processed precision reduced spectral domain representation 131 , may be considered a lossless signal coding chain and the signal coding chain shown on the bottom of Fig. 1a, e.g. resulting in the residual information, may be considered a potentially lossy signal coding chain.

[0235] Hence, the combination of both these signal coding chains in Fig. 1a enables a scalable and precisely-controlled (e.g. with respect to an influence of the filtering / prediction to quantization noise) near lossless signal coding chain.

[0236] Optionally, the decoded version of the residual information 121 may represent a less significant, e.g. low-order, portion of a number representation of a spectral domain representation of the audio signal, and the filtered and / or prediction processed precision reduced spectral domain representation 131 may represent a more significant, e.g. high-order, portion of the number representation of the spectral domain representation of the audio signal. Accordingly, the combined spectrum 141 may represent (or represent an approximation of) the number representation, e.g. the envisioned number representation, of the spectral domain representation of the audio signal.

[0237] In line with the above embodiments, the decoder 100 may be configured to derive or determine the filtered and / or prediction processed precision reduced spectral domain representation 131 in the form of a first number representation and to derive or determine the decoded version of the residual information 121 as a second number representation, so as to combine said

[0238] FH240905PEP-2025315045. DOCX number representations using the combining unit 140, to obtain the number representation of the spectral domain representation of the audio signal.

[0239] For example, the first number representation may be a first portion of the number representation of the spectral domain representation of the audio signal, e.g. a set of most significant bits (MSBs), and the second number representation may be a second portion of the number representation of the spectral domain representation of the audio signal, e.g. a set of less significant bits (LSBs).

[0240] As an example, the decoder 100, e.g. in particular the combining unit 140, may be configured to concatenate or to combine, the first number representation and the second number representation, in order to obtain the number representation of the spectral domain representation of the audio signal. The combining unit 140 may hence, be configured to allocate predetermined bit positions of the number representation of the spectral domain representation of the audio signal to the respective bits of the first and second number representation.

[0241] As an example, the decoded version of the residual information 121 may represent one or more least significant bits, LSBs, of the number representation of the spectral domain representation of the audio signal, and the filtered and / or prediction processed precision reduced spectral domain representation 131 may represent one or more most significant bits, MSBs, of the number representation of the spectral domain representation of the audio signal.

[0242] For example, a bitstream according to embodiments may hence comprise the encoded representation 101 of the precision reduced spectral domain representation of the signal, the encoded representation 102 of a residual information and optionally additionally an information describing a splitting between the representation of the precision reduced spectral domain representation of the signal and the representation of the residual information, e.g. as a side information, see e.g. 202 in Fig. 2.

[0243] Hence, the bitstream may comprise an information of how the (e.g. original) spectrum is split into the precision reduced and the residual representation.

[0244] As an optional feature, the information included in the encoded representation of the residual information 102, may be a scalable information, hence comprising more or less detailed information about the residual and hence enabling a more or less accurate reconstruction of the original audio signal. Hence, the decoder 100 may optionally be configured to determine

[0245] FH240905PEP-2025315045. DOCX the decoded version 121 of the residual information with variable accuracy, for example depending on an amount (e.g. a number of encoded bits) of residual information included in the encoded representation of the residual information 102. Hence, decoding unit 120 may be configured to decode encoded representations 102 with variable bit length. This may improve the scalability of the coding scheme.

[0246] Furthermore, e.g. in particular in combination with the above example, the combining unit 140 may be configured to handle such variable numbers of encoded bits for the reconstruction of the number representation of the combined spectrum 141.

[0247] For example, the decoder 100 may be configured to handle cases in which a number of least significant bits included in the encoded representation 102 of the residual information is smaller than expected in view of a significance (e.g. bit weight) of the more significant portion of a number representation of the spectral domain representation. For example, the decoder may be configured to handle cases in which the more significant portion of a number representation of the spectral domain representation is fully encoded and in which one or more associated least significant bits are not encoded within the encoded representation 102 of the residual information, e.g. in order to save bit rate.

[0248] As an example, the decoder may be configured to fill up “missing” bits, e.g. with default values. Bits may be determined as missing based on the bit weight. As an example, if the number representation of the spectral domain representation has 5 digits with MSBs defining digits 3 to 5 and LSBs defining digits 1 to 2, the decoder may adapt the determination of the number representation of the spectral domain representation, if instead of the 2 expected LSBs (e.g. expected because of the 5 total digits only the 3 MSBs were received), less than 2 LSBs or even no LSBs were received, e.g. so as to fill up missing LSBs with default values to approximate the number representation of the spectral domain representation with the received MSBs and the default values.

[0249] Here, reference is made to Fig. 1 b. Fig. 1 b shows a schematic view of a decoder 100’ for decoding a signal using a prediction information according to an embodiment.

[0250] Decoder 100’ is configured to decode an encoded representation 10T of a prediction information (e.g. using decoding unit 110’), in order to obtain a decoded version of the prediction information 11 T. Furthermore, the decoder 100’ is configured to decode an encoded representation 102 of a residual information (e.g. using decoding unit 120), in order to obtain a decoded version of the residual information 121.

[0251] FH240905PEP-2025315045. DOCX In addition, the decoder is configured to obtain a prediction processed precision reduced spectral domain representation 13T on the basis of the prediction information 111’ using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, here, as an example using prediction processing unit 130’.

[0252] Furthermore, the decoder 100’ is configured to determine a combined spectrum 14T based on the prediction processed precision reduced spectral domain representation and based on the decoded version of the residual information, for example using combining unit 140.

[0253] Hence, in comparison to Fig. 1a, the decoder of Fig. 1b may obtain a prediction information, which may, for example, be not limited to or may, for example, not comprise an encoded representation of a (e.g. precision reduced) spectral domain representation (which my, for example, be a prediction processing residual), but may for example in addition or, for example, solely comprise prediction parameters, (e.g., input vector or input data to be processed, for example, as input by a DNN performing a prediction processing, e.g., control information for a DNN performing a prediction processing, e.g., a prediction processing residual, e.g., an excitation signal to be processed as input by a prediction processing structure, e.g., input vector or input data to be processed as input by a prediction processing structure, e.g., prediction coefficients), for performing a prediction processing, e.g. a prediction, using the information to obtain the prediction processed precision reduced spectral domain representation 131’.

[0254] As an example, in contrast, the precision reduced spectral domain representation 111 of Fig. 1a may for example comprise or may, for example, solely be a prediction residual based on which, e.g. in combination with prediction parameters (which, the decoder 100 may optionally obtain, e.g. in the form of an encoded representation thereof), a prediction processed precision reduced spectral domain representation may be obtained.

[0255] Hence, the prediction processed precision reduced spectral domain representation may, for example, be obtained losslessly.

[0256] Next, reference is made to Fig. 2. Fig. 2 shows a schematic view of a decoder with additional optional features according to embodiments of the invention. The decoder 200 comprises the previously discussed units of Fig. 1 a. These units may have the same or a similar functionality, as well as the same or similar optional features and functionalities as discussed before.

[0257] FH240905PEP-2025315045. DOCX In addition, as an optional feature, decoder 200 comprises a decoding unit 210 (which may be again distinct from or similar to decoding units 110 and 120 or may even be part of a common decoding unit replacing all three decoding units 110, 120 and 210). The decoder 200 is configured to obtain an optional encoded representation of a filtering / prediction processing information 201 , which may be decoded, and based on the decoded version thereof 211 , the filtering and / or prediction processing may be adjusted.

[0258] As an example, the filtering and / or prediction information may comprise a set of filter parameters and / or prediction processing parameters. For example, information 211 may also comprise a prediction filter information for adjusting a prediction filtering unit 130.

[0259] As an example, the decoder 200 may be configured to effect a temporal noise shaping and / or achieve a prediction gain in dependence on a set of filter parameters and / or in dependence on a set of prediction processing parameters, included in the filtering and / or prediction processing information 211.

[0260] For example, the filtering and / or prediction processing information 211 may alternatively or in addition comprise a mode flag, indicating whether a rounding operation of the filter and / or prediction processing should be activated or deactivated. However, it is to be noted that such an information about a respective mode flag may as well be provided in the form of a separate side information.

[0261] As an example, the decoder 200 may be configured to adapt a filtering performed using the filter structure (e.g. as an example of unit 130) based on the filter parameters (e.g. included in information 211), in order to adapt a weighting of a prediction of the filtering (e.g. in case the filter is a prediction filter), and / or to a adapt a prediction order of the filtering (e.g. in case the filter is a prediction filter) and / or in order to reconfigure the filter structure.

[0262] As an example, the decoder may be configured to adapt the prediction processing using the prediction processing structure (e.g. as an example of unit 130) based on the prediction processing parameters, in order to adapt a weighting of the prediction, and / or to a adapt a prediction order of the prediction and / or in order to reconfigure the prediction processing structure. Hence, the coding efficiency may be increased by such a flexible adaptation of the filtering / prediction unit 130.

[0263] FH240905PEP-2025315045. DOCX Furthermore, as an optional feature, the decoder 200 may be configured to adapt based on the filter and / or prediction processing parameters (e.g. included in information 211) whether a filtering and / or a prediction processing is performed in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction.

[0264] Alternatively or in addition, the decoder 200 may be configured to adapt for at least one filtering direction and / or for at least one combination of filtering directions the filtering performed using the filter structure (see unit 130) based on the filter parameters, in order to adapt a weighting of a prediction of the filtering, and / or to a adapt a prediction order of the filtering and / or in order to reconfigure the filter structure and / or in order to change filter coefficients of the filter structure.

[0265] Alternatively or in addition, the decoder 200 may be configured to adapt for at least one prediction processing direction and / or for at least one combination of prediction processing directions the prediction processing performed using the prediction processing structure (see unit 130) based on the prediction processing parameters, in order to adapt a weighting of the prediction processing, and / or to adapt a prediction order of the prediction processing and / or in order to reconfigure the prediction processing structure and / or in order to change prediction processing coefficients of the prediction processing structure

[0266] As an optional feature, the decoded version of the precision reduced spectral domain representation 111 may comprise integer-valued spectral bin values for a plurality of spectral bins.

[0267] Furthermore, the decoder may be configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation 131 in the form of an integer valued filtered and / or prediction processed precision reduced spectral domain representation.

[0268] Hence, optionally, the filtering / prediction processing unit may be configured to perform a rounding operation. For example, the decoder 100, 200 may be configured to map integer values of the decoded version 111 of the precision reduced spectral domain representation onto integer values of the filtered and / or prediction processed precision reduced spectral domain representation 131 using a rounding of prediction processing or filtered or prediction filtered values.

[0269] Optionally, the decoded version 111 of the precision reduced spectral domain representation of the signal may, for example, be an energy-reduced version and / or redundancy-reduced

[0270] FH240905PEP-2025315045. DOCX version and / or version where predictable signal portions have been removed of a precision reduced spectral domain representation of the signal, and the decoder 100, 200 may be configured to losslessly reconstruct the precision reduced spectral domain representation of the signal using the filtering and / or using the prediction processing.

[0271] As an optional feature, the filtering and / or prediction processing information may comprise a filter mode information and / or prediction processing mode information. Hence, the decoder 100, 200 may be configured to adapt the filtering and / or prediction processing of the precision reduced spectral domain representation 111.

[0272] Based on the mode information (e.g. in particular prediction processing mode and / or filtering mode information) the decoder, 100, 200 may, for example, selectively activate or deactivate a rounding functionality of the filtering and / or prediction processing unit 130.

[0273] As an optional feature, the filtering / prediction information 211 may, for example, comprise an information describing the operation mode of the encoder used to encode the encoded representation of the precision reduced spectral domain representation. Based on the operating mode information, e.g. included in information 211 , the decoder may be configured to adapt the filtering and / or prediction processing of the precision reduced spectral domain representation 111. Alternatively or in addition (e.g. based on the operating mode information), the decoder may optionally be configured to selectively determine the combined spectrum 141 based on the filtered and / or prediction processed precision reduced spectral domain representation 131 and the decoded version of the residual information 121 or based on the filtered and / or prediction processed precision reduced spectral domain representation 131 without considering a residual information.

[0274] Optionally, the above-discussed additional information such as an information about the operating mode may, for example, be provided to the combining unit 140, e.g. based on a respective decoding of a respective additional encoded information.

[0275] Furthermore, based on such an encoder mode information, the decoder may set itself accordingly, e.g. to achieve a desired coding scheme, e.g. in the form of transformation (e.g. IntMDCT) + filtering / prediction processing (e.g. TNS) (encoder-sided) and inverse filtering / prediction processing (e.g. inverse TNS) and inverse transformation (e.g. IntMDCT) (decodersided).

[0276] FH240905PEP-2025315045. DOCX Moreover, as an example, the decoder 200 may be configured to use a default residual information 242 instead of the decoded version of the residual information 121 , for example, if the encoded representation 102 of the residual information is absent (or for example a portion thereof) in a respective bitstream (e.g. enabling a further reduction of used transmission resources).

[0277] As an optional feature, on the basis of an internal decision mechanism (e.g. using the filtering and / or prediction processing information 211), the decoder 100, 200, may be configured to selectively enable or disable usage of a rounding functionality in the filtering in the frequency direction and / or in the time direction, and / or in the spatial direction and / or in the channel direction; and / or to selectively enable or disable usage of a rounding functionality in the prediction processing in the frequency direction and / or in the time direction, and / or in the spatial direction and / or in the channel direction.

[0278] This may, for example, enable a switching between a lossless or near lossless spectrum reconstruction (e.g. of the filtered precision reduced spectral domain representation) and a lossy spectrum reconstruction (e.g. of the spectral domain representation).

[0279] Hence, optionally, the filtering and / or prediction processing in the frequency direction and / or in the time direction, and / or in the spatial direction and / or in the channel direction may be configured to perform a lossless reconstruction of the filtered and / or prediction processed precision reduced spectral domain representation.

[0280] As an example, the unit 130 may be a TNS, temporal noise shaping, filter with lossless reconstruction (e.g. enabling a lossless reconstruction), configured to perform a filtering in the frequency direction.

[0281] As an optional feature, decoder 200 may be configured to obtain or to receive a side information 202. The side information 202 may describe a splitting between the representation of the precision reduced spectral domain representation of the signal and the representation of the residual information.

[0282] Optionally, the side information 202 may be included in the bit stream comprising the encoded representation 101 of a precision reduced spectral domain representation of the signal and the encoded representation 102 of the residual information. Hence, optionally, the side information 202 may be obtained as an encoded representation and may be decoded as well (not shown).

[0283] FH240905PEP-2025315045. DOCX The side information 202, e.g. via bitstream or from another source, may as well comprise an information whether a rounding should be applied in the filtering and / or prediction processing in the frequency direction and / or in the time direction, and / or in the spatial direction and / or in the channel direction. Optionally, the side information 202 may comprise the filtering / prediction processing information.

[0284] As an optional feature, the side information 202 may, for example, comprise or be an operating mode information. The decoder 100, 200 may, for example, be configured to switch between a lossless mode, a near lossless mode and a lossy mode, e.g. in dependence on an operation mode information, e.g. a flag.

[0285] The decoder may hence be configured to decode and to losslessly, e.g. using a rounding, filter and / or predict a spectral domain representation, e.g. a complete representation of spectral domain values, if the decoder is in the lossless mode.

[0286] The decoder may hence further be configured to decode and to losslessly, e.g. using a rounding, filter and / or predict the precision reduced spectral domain representation (and to supplement then the filtered and / or prediction processed precision reduced spectral domain representation using the residual spectral domain representation 121 as far as an encoded residual spectral domain representation 102 is received by the decoder), if the encoder is in the near lossless mode.

[0287] Furthermore, such a decoder, e.g. 100, 200, may be configured to decode and to lossily, e.g. without using a rounding, filter and / or predict a spectral domain representation, e.g. the precision reduced spectral domain representation, if the encoder is in the lossy mode.

[0288] Hence, in other words, given the operating mode information indicates a lossless operating mode, for example, no residual information may be provided to the decoder. Based on the side information 203 comprising the operating mode information, the filtering / prediction processing unit may be adjusted so as to perform a rounding operation and no combining may be performed. The combined spectrum 141 may hence correspond to the filtered and / or prediction processed reduced spectral domain representation, wherein the filtered and / or prediction processed precision reduced spectral domain representation 131 is in this case a predicted full spectral domain representation without precision reduction, since in the lossless operating mode, the filtering / prediction processing unit may act as a predictor achieving a prediction gain.

[0289] FH240905PEP-2025315045. DOCX Hence, in other words, given the operating mode information indicates a lossy operating mode, for example, also no residual information may be provided to the decoder. Based on the side information 203, comprising the operating mode information, the filtering / prediction processing unit may be adjusted so as to perform no rounding operation and no combining may be performed. The combined spectrum 141 may hence correspond to the filtered and / or prediction processed precision reduced spectral domain representation 131 , wherein the filtered and / or prediction processed precision reduced spectral domain representation 131 is in this case a perceptually filtered lossy version of the original spectral domain representation, since in the lossy operating mode, the filtering / prediction processing unit may act as a noise shaping filter, e.g. a temporal noise shaping filter.

[0290] Hence, in other words, given the operating mode information indicates a near lossless operating mode, the residual information may be provided to the decoder. Based on the side information 203 comprising the operating mode information, the filtering / prediction processing unit 130 may be adjusted so as to perform a rounding operation to enable a lossless signal coding chain for the filtered and / or prediction processed precision reduced spectral domain representation 131 (e.g. realizing a prediction gain, e.g. wherein the filtered and / or prediction processed precision reduced spectral domain representation 131 may, for example, be a prediction processed precision reduced spectral domain representation 131). The combined spectrum 141 may hence correspond to the combination of the losslessly coded filtered and / or prediction processed precision reduced spectral domain representation 131 and the potentially lossily coded residual information 121.

[0291] As an optional feature, a structure and / or parametrization of the filtering / prediction processing unit 130 may, for example, remain unchanged irrespective of the operating mode (e.g. at least irrespective of whether the decoder is an operating mode corresponding to an encoder sided redundance reduction operating mode or in a noise shaping operating mode).

[0292] Optionally, decoders 100 and 200 may comprise a transform unit, configured to retransform the combined spectrum, e.g. to a time-domain signal.

[0293] Next, reference is made to Fig. 3a. Fig. 3a shows a schematic view of an encoder for encoding a signal according to embodiments of the invention. Encoder 300 comprises a transform unit 310, a precision reduction unit 320, a residual determination unit 330, a filtering / prediction processing unit 340 (e.g. a filtering unit, e.g. a prediction processingunit, e.g. a prediction filtering unit, e.g. a predictive filtering unit; wherein the filtering / prediction processing unit 130 may, for example, be considered an inverse version of the unit 340, e.g. performing an inverse

[0294] FH240905PEP-2025315045. DOCX processing or performing inverse mathematical operations) and an encoding unit 350. Encoder 300 is configured to obtain signal 301 (e.g. an audio signal), and to derive, e.g. using transform unit 310, a spectral domain representation 311 thereof. Alternatively, the spectral domain representation 311 may be provided directly to encoder 300. This spectral domain representation 311 may be considered an initial or an original spectral domain representation, e.g. a spectral domain representation in a full integer value representation, for example in the form of a MDCT spectrum.

[0295] Furthermore, encoder 300 is configured to obtain, e.g. using precision reduction unit 320, a precision reduced spectral domain representation 321 of the signal based on the spectral domain representation 311. For example, precision reduction unit 320 may be configured to use or apply a rounding operation, e.g. using a filtering comprising a rounding operation. For example, the precision reduced spectral domain representation may be obtained in the form of a more significant portion of a number representation of the spectral domain representation of the audio signal, e.g. in the form of a first number representation, e.g. in the form of most significant bits, MSBs, of the number representation of the spectral domain representation of the audio signal.

[0296] Furthermore, encoder 300 is configured to obtain a residual information 331 based on the spectral domain representation 311. Optionally, e.g. as indicated in Fig. 3a, the encoder may be configured to determine the residual information 331 as a residual spectral domain representation or a portion thereof, for example as a residual between the spectral domain representation 311 and the precision reduced spectral domain representation 321 , hence, for example, representing a difference between the spectral domain representation 311 and the precision reduced spectral domain representation 321 (which may hence optionally be provided to residual determination unit 330).

[0297] The residual information 331 may, for example, be obtained in the form of a less significant, e.g. low-order, portion of a number representation of a spectral domain representation, e.g. in the form of a second number representation, e.g. for the less significant portion of the number representation of the spectral domain representation of the audio signal, e.g. in the form of one or more least significant bits, LSBs, of the number representation of the spectral domain representation of the audio signal.

[0298] Furthermore, the encoder 300 is configured to obtain a filtered and / or prediction processed precision reduced spectral domain representation 341 on the basis of the precision reduced spectral domain representation 321 using a filtering in a frequency direction and / or in a time

[0299] FH240905PEP-2025315045. DOCX direction and / or in a spatial direction and / or in a channel direction, and / or using a prediction processingin a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction. Moreover, unit 340 may optionally comprise, or be, a prediction filter (or a predictive filter) for using a prediction filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction for obtaining the filtered and / or prediction processed precision reduced spectral domain representation 341.

[0300] In particular, as an example, the filtering / prediction processing unit 340 may, for example comprise or be a temporal noise shaping filter structure, in order to effect a prediction filtering in a frequency direction of the precision reduced spectral domain representation 321.

[0301] In addition, encoder 300 is configured, e.g. using encoding unit 350, to obtain an encoded representation 351 on the basis of the filtered and / or prediction processed precision reduced spectral domain representation and the residual information 331.

[0302] Optionally, the spectral domain representation 311 may be an integer spectral domain representation of the signal 301 and / or the encoder 300 may be configured to determine the spectral domain representation 311 as an integer spectral domain representation (e.g. as an Integer MDCT (e.g. IntMDCT) spectrum or a portion thereof, e.g. using a rounding).

[0303] Hence, the encoder 300 may optionally be configured to obtain, e.g. to determine, e.g. to calculate, the precision reduced spectral domain representation 321 of the signal in the form of an integer precision reduced spectral domain representation based on the integer spectral domain representation.

[0304] Accordingly, as an optional feature, the encoder may be configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation 341 as a filtered and / or prediction processed integer precision reduced spectral domain representation on the basis of the integer precision reduced spectral domain representation 321 using a filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and / or using a prediction processing (e.g. a respective encoder-sided prediction processing)in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction.

[0305] As an optional feature the encoder 300 may be configured to adapt the filtering and / or prediction processing in frequency direction and / or in time direction and / or in spatial direction and / or in channel direction based on the available bitrate. For example, the encoder 300 may

[0306] FH240905PEP-2025315045. DOCX determine such an available bitrate, or may be provided with an information about the same and may adjust the filtering / prediction processing unit 340 accordingly.

[0307] As an example, the encoder 300 may be configured to selectively use a filtering and / or prediction processing with rounding or without rounding. Hence, filtering / prediction processing unit 340 may be a switchable filtering / prediction unit 340, e.g. having an operating mode with and without rounding. Optionally, the precision reduction unit 320 may, for example, also comprise operating modes with or without rounding and the encoder may, for example, be configured to determine the precision reduced spectral domain representation 321 of the signal selectively with or without rounding.

[0308] Hence, encoder 300 may be configured to use a filtering or prediction processing or prediction filtering with rounding, if an available bitrate is sufficient to losslessly encode the filtered and / or prediction processed precision reduced spectral domain representation 341 and to otherwise use a filtering or prediction processing or prediction filtering without rounding.

[0309] Here, reference is made to Fig. 3b. Fig. 3b shows a schematic view of an encoder 300’ for encoding a signal using a prediction information according to an embodiment. Encoder 300’ is configured to obtain a spectral domain representation of the signal 311 and to obtain a precision reduced spectral domain representation 321 of the signal based on the spectral domain representation, In addition encoder 300’ is configured to obtain a residual information 331 based on the spectral domain representation.

[0310] Hence, regarding the before-discussed elements, encoder 300’ may have same or corresponding functionalities than encoder 300.

[0311] However, for example as a generalization or in contrast to encoder 300, encoder 300’ is configured to obtain a prediction information 341’ on the basis of the precision reduced spectral domain representation using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction. Furthermore, the encoder 300’ is configured to obtain an encoded representation 351’ on the basis of the prediction information and on the basis of the residual information.

[0312] In line with the discussion of Fig. 1b, encoder 300’ may be configured to obtain the precision reduced spectral domain representation of the signal 321 so as to encode the same residual- free, e.g. so that the prediction information 341’ only comprises prediction parameters, such as neural network parameters. In other words, the prediction information may, for example

[0313] FH240905PEP-2025315045. DOCX represent a residual free and energy reduced or redundancy reduced version of the precision reduced spectral domain representation.

[0314] However, it is to be noted that the embodiments according to Fig. 1b and 3b may, for example, also comprise the features, functionalities and details of the embodiments discussed in Fig. 1a, 2, 3a and 4. In particular, the encoder 300’ may, for example as well be configured to provide a prediction information 34T, which may comprise prediction parameters and a prediction residual, which may be considered, e.g. in combination with the prediction parameters or alone, as an prediction processed precision reduced spectral domain representation.

[0315] Next, reference is made to Fig. 4, showing a schematic view of an encoder with additional optional features, according to embodiments of the invention. Encoder 400 comprises features of encoder 300 as discussed previously.

[0316] Furthermore, as an optional feature, the encoder 400 may be configured to obtain, for example to receive, for example to determine, a filtering / prediction processing information 442. This information 442 may, for example, comprise an information about the available bitrate, for example, to switch between a filtering and / or prediction processing operating mode with or without rounding. Such an information may, for example, be received by encoder 400 from an external source or may be determined (e.g. decided) internally.

[0317] An information about the available bitrate may optionally, also be provided to the transform unit 310 (e.g. to switch between obtaining an MDCT or an IntMDCT), the precision reduction unit 320 (e.g. to switch between an operating with or without rounding) and / or the encoding unit 350.

[0318] In addition or alternatively, information 442 may, for example, comprise a set of filter parameters and / or a set of prediction processing parameters and / or a set of prediction filter parameters, for use with the filtering / prediction processing unit 340. Hence, the properties of the filtering and / or prediction processing may be adapted selectively.

[0319] As indicated as an example, such an information 442 may be provided to the filtering / prediction processing unit 340.

[0320] For example, the filtering / prediction processing unit 340 may comprise a temporal noise shaping filter structure and the filtering / prediction processing information 442 may comprise

[0321] FH240905PEP-2025315045. DOCX temporal noise shaping filter parameters and an information about an available bitrate. Hence, encoder 400 may adjust the unit 340 (e.g. filter unit), so as to achieve a prediction gain in dependence on the set of filter parameters.

[0322] Moreover, the encoder may, for example, be configured to use a temporal noise shaping filtering with rounding, if an available bitrate is sufficient to losslessly encode the filtered and / or prediction processed precision reduced spectral domain representation 341 obtainable by the temporal noise shaping filtering with rounding, and to use the temporal noise shaping filtering without rounding, if the available bitrate is insufficient to losslessly encode the filtered and / or prediction processed precision reduced spectral domain representation 341 obtainable by the temporal noise shaping filtering with rounding.

[0323] As an optional feature, encoder 400 may, for example, be configured to encode, e.g. in the bitstream, a filter mode information 443, indicating whether a filtering with rounding or a filtering without rounding was performed, in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation 341 .

[0324] Optionally, the encoder may, be configured to encode a prediction processing mode information 443 indicating whether a prediction processing with rounding or a prediction processing without rounding was performed in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation 341.

[0325] Such an encoded representation of the mode information may, for example, be part of the encoded representation 351 or may, for example, be an additional side information in a respective bitstream. Information 443 may, for example, be included in or may even be the side information 202.

[0326] As an optional feature, encoder 300, 400 may, for example, be configured to perform the filtering and / or the prediction processing to obtain the filtered and / or prediction processed precision reduced spectral domain representation 341 independent from the residual information 331.

[0327] In general, encoder 300, 400 may, for example, be configured to switch between different operating modes, e.g. in dependence on an available bitrate. As discussed previously, the encoder may determine such a bitrate information itself, or may be provided with the same.

[0328] FH240905PEP-2025315045. DOCX In particular, as an optional feature, based on an available bitrate, the encoder 300, 400, may be configured to switch between a lossless mode, a near lossless mode and a lossy mode.

[0329] For example, the encoder 400 may be configured to obtain and to encode a filtered and / or prediction processed version, e.g. a losslessly filtered version, of the spectral domain representation (see e.g. 444), if the encoder is in the lossless mode. For example, signal 311 may bypass the precision reduction unit 320 and may be provided to the filtering / prediction unit 340 directly. The encoder 400 may operate the filtering / prediction processing unit 340 in a lossless filtering / prediction mode, e.g. using invertible rounding operations (e.g. so that a corresponding decoder can losslessly reconstruct the spectral domain representation 311) and / or using integer processing.

[0330] As an example, in the lossless mode, the encoder 400 may be configured to perform a filtering in frequency direction with rounding and / or a filtering in time direction with rounding and / or a filtering in spatial direction with rounding and / or a filtering in channel direction with rounding, and / or a prediction processing in frequency direction with rounding and / or a prediction processing in time direction with rounding and / or a prediction processing in spatial direction with rounding and / or a prediction processing in channel direction with rounding of the spectral domain representation 311 in order to obtain a filtered version of the, e.g. full, spectral domain representation in the form of an integer representation and to encode the filtered version of the spectral domain representation (see e.g. 444).

[0331] For example, the encoder may be configured to encode the filtered and / or prediction processed precision reduced spectral domain representation 341 or at least a part thereof, if the encoder is in the near lossless mode (e.g. as shown together with the residual information, e.g. a scalable amount of residual information, e.g. based on an available bitrate).

[0332] As an example, in the near lossless mode, the encoder 400 may be configured to perform a filtering in frequency direction with rounding and / or a filtering in time direction with rounding and / or a filtering in spatial direction with rounding and / or a filtering in channel direction with rounding, and / or a prediction processing in frequency direction with rounding and / or a prediction processing in time direction with rounding and / or a prediction processing in spatial direction with rounding and / or a prediction processing in channel direction with rounding of the precision reduced spectral domain representation 321 , in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation 341 in the form of an integer representation and to encode the filtered and / or prediction processed precision reduced spectral domain representation, or at least a part thereof.

[0333] FH240905PEP-2025315045. DOCX For example, the encoder may be configured to obtain and to encode a filtered and / or prediction processed version of the spectral domain representation (see e.g. 444) or at least a part thereof, if the encoder is in the lossy mode. Therefore, the spectral domain representation may, for example, bypass the precision reduction unit 320 and may be provided directly to the filtering / prediction unit 340. The filtering / prediction processing unit may, for example apply a lossy filtering and / or prediction processing.

[0334] As an example, in the lossy mode, the encoder may be configured to perform a filtering in frequency direction without rounding and / or a filtering in time direction without rounding and / or a filtering in spatial direction without rounding and / or a filtering in channel direction without rounding, and / or a prediction processing in frequency direction without rounding and / or a prediction processing in time direction without rounding and / or a prediction processing in spatial direction without rounding and / or a prediction processing in channel direction without rounding of the spectral domain representation 311 , in order to obtain a filtered version of the spectral domain representation and to encode the filtered version of the spectral domain representation, or at least a part thereof.

[0335] As an example, the encoder 400 may, for example, be configured to encode an information about the operating mode of the encoder. This information may be part of the encoded representation 351 or may, for example be a separate bitstream element.

[0336] Optionally, the encoder 400 may be configured to determine the precision reduced spectral domain representation 320 in dependence on an available bitrate and / or in dependence on the spectral domain representation 311. Hence, as an example, the precision reduction unit 320 may, for example, be provided with the information about the available bitrate.

[0337] Hence, the encoder 400 may, for example, be configured to determine the precision reduced spectral domain representation 321 , and consequently the filtered and / or prediction processed precision reduced spectral domain representation 341 , in order to losslessly transmit the filtered and / or prediction processed precision reduced spectral domain representation.

[0338] As an example, the encoder 400 may, for example, be configured to select a number of bits representing the precision reduced spectral domain representation 321 (and possibly also the number of bits representing the residual information 331 , which may be “complementary” to the number of bits representing the precision reduced spectral domain representation) in dependence on the available bitrate.

[0339] FH240905PEP-2025315045. DOCX Hence, optionally, the encoder 400 may be configured to determine a bit representation (e.g. a binary representation; e.g. a m-ary representation with m>=2) of the filtered and / or prediction processed precision reduced spectral domain representation 341 and to losslessly encode said bit representation, e.g. using encoding unit 350.

[0340] Optionally, the encoder 400 may, for example, be configured to determine a bit representation of the residual information 331 , e.g. using residual determination unit 330, and to selectively encode said bit representation, e.g. in full, a portion of said bit representation, e.g. a true subset of bits of said bit representation or to selectively not encode said bit representation in dependence on an available bitrate. Hence, encoding unit 350 may, for example, be provided with the information about the available bitrate, in order to determine how much information about the residual information is to be encoded, e.g. in which precision the residual information is to be encoded.

[0341] For example, the encoder 400 may, for example, comprise a dedicated functionality for encoding the residual information 331 , wherein said functionality is configured to adapt a size of the residual information 331 or an amount of residual information that is encoded.

[0342] As an example, the encoding unit 350 may, for example, be configured to encode the filtered and / or prediction processed precision reduced spectral domain representation 341 and the residual information 331 separately in a bitstream.

[0343] Here again, it is to be noted that embodiments enable creating two distinct signal encoding chains, which allow for a precise control of the influences of filtering / prediction processing and quantization noise, e.g. independent from one another, to distinct portions of the spectral domain representation 311 of the signal 301. Based on a splitting or approximation of the spectral domain representation of the signal into the precision reduced spectral domain representation 321 and the residual information 331 (wherein the latter may only enable an approximation of the original spectral domain representation and not a perfect reconstruction), the filtering / prediction processing may be limited to a signal portion 321 , a lossless reconstruction of which can, under some constraints, be guaranteed in a respective decoder. Hence, an undesired effect of the filtering / prediction processing (e.g. in high prediction gain cases) may be avoided.

[0344] FH240905PEP-2025315045. DOCX This is particularly advantageous in near lossless operating modes, wherein high prediction gains can be achieved, but wherein such high prediction gains would lead in conventional concepts to undesired artifacts because of strong amplifications of quantization noise.

[0345] However, it was recognized that based on the two signal paths of inventive encoders, this problem can be solved.

[0346] Furthermore, the more “definitive” operating modes, such as fully lossless or fully lossy can be simply covered by bypassing the precision reduction unit 320 (e.g. defining the full original spectrum 311 as MSBs) and not determining a residual information 331 and by adjusting the filtering / prediction processing accordingly (e.g. in an invertible manner, e.g. with rounding).

[0347] Next, reference is made to Fig. 5. Fig. 5 shows a schematic view of another encoder according to embodiments of the invention. Encoder 500 comprises a transform unit 310, configured to obtain, e.g. derive a spectral domain representation 311 of a signal 301 . Alternatively, encoder 500 may be configured to receive the spectral domain representation 311.

[0348] Furthermore, encoder 500 comprises a switching unit 510, which is configured to provide a filtering / prediction processing unit 340 with the spectral domain representation 311 or with a precision reduced version thereof, 321 , depending on an operating mode information 511.

[0349] The encoder 500 is configured to obtain a precision reduced spectral domain representation 321 of the signal based on the spectral domain representation 311 , using precision reduction unit 320.

[0350] The encoder 500 is configured to obtain a filtered and / or prediction processed spectral domain representation, see 444, on the basis of the spectral domain representation using the filtering and / or prediction processing unit 340, and to obtain an encoded representation 351” on the basis of the filtered and / or prediction processed spectral domain representation, if the encoder 500 is in the noise shaping operating mode.

[0351] Furthermore, the encoder 500 is configured to obtain a filtered and / or prediction processed precision reduced spectral domain representation, see 341 , on the basis of the precision reduced spectral domain representation 321 using the filtering and / or prediction processing structure and to obtain an encoded representation 351” on the basis of the filtered and / or prediction processed precision reduced spectral domain representation 321 if the encoder 500 is in the redundancy reduction operating mode.

[0352] FH240905PEP-2025315045. DOCX As discussed in the context of Fig. 1 to 4, the filtering / prediction processing unit may, for example, be or may, for example, comprise a filter structure, a prediction processing structure and / or a prediction filter or predictive filter structure.

[0353] Units having same reference numbers as discussed in Fig. 3 and 4 may have same or similar functionalities details and features, as those shown in Fig. 5.

[0354] As an optional feature, the encoder 500 is configured to obtain the filtered and / or prediction processed spectral domain representation, see 341 , in the noise shaping operating mode and the filtered and / or prediction processed precision reduced spectral domain representation, see 341 , in the redundancy reduction operating mode using the same filter structure and filter parameters, which are independent of the whether the encoder is in the noise shaping operating mode or in the redundancy reduction operating mode and / or using the same prediction processing structure and prediction processing parameters, which are independent of whether the encoder is in the noise shaping operating mode or in the redundancy reduction operating mode.

[0355] Moreover, optionally, irrespective of the operating mode, the filtering and / or prediction processing performed in filtering / prediction unit 340 may, for example, be performed in a same manner. Hence, the encoder 500 may, for example, be configured to obtain the filtered and / or prediction processed spectral domain representation in the noise shaping operating mode and the filtered and / or prediction processed precision reduced spectral domain representation in the redundancy reduction operating mode using the same filtering and / or using the same prediction processing.

[0356] Next, reference is made to Fig. 6. Fig. 6 shows a schematic view of another encoder with additional, optional features, according to embodiments of the invention. Compared to encoder 500, encoder 600 comprises a residual determination unit 330. Accordingly, the encoder 600 may, for example, be configured to obtain a residual information 331 based on the spectral domain representation and to obtain an encoded representation 351” on the basis of the filtered and / or prediction processed precision reduced spectral domain representation and on the basis of the residual information, if the encoder is in the redundancy reduction operating mode.

[0357] Units having same reference numbers as discussed in Fig. 3 and 4 may have same or similar functionalities details and features.

[0358] FH240905PEP-2025315045. DOCX As an example, the encoder 600, may, for example, be configured to switch between the operating modes in dependence on an available bitrate. For example, the operating mode information 511 may be determined by such an information about the available bitrate.

[0359] The encoder 600 may, for example comprise at least three operating modes, namely a lossless mode, a near lossless mode and a lossy mode. The switching may, for example, be performed based on the available bitrate.

[0360] Accordingly, the encoder 600 may be configured to obtain and to encode a filtered and / or prediction processed version, e.g. a losslessly filtered version, e.g. using unit 340, of the spectral domain representation 311 , if the encoder is in the lossless mode, the encoder 600 may be configured to obtain and to encode the filtered and / or prediction processed precision reduced spectral domain representation or at least a part thereof, e.g. the losslessly filtered, precision reduced spectral domain representation or a part thereof, if the encoder is in the near lossless mode, and the encoder 600 may be configured to obtain and to encode a filtered and / or prediction processed version, e.g. a lossily filtered version, of the spectral domain representation 311 or at least a part thereof, if the encoder is in the lossy mode.

[0361] Hence, the operating mode information 511 or respectively an information about an available bitrate may, for example, be provided to the filtering / prediction unit 340 in order to set a specific filtering or prediction mode of the unit (e.g. with or without rounding).

[0362] Accordingly, as an example, the encoders 600 may be configured to switch at least between a lossy version of the noise shaping operating mode, a near lossless version of the noise shaping operating mode, a near lossless version of the redundancy reduction operating mode, and a lossless version of the redundancy reduction operating mode, e.g. in dependence on an available bitrate.

[0363] Therefore, optionally, the encoder 600 may be configured to perform a filtering in frequency direction with rounding and / or a filtering in time direction with rounding and / or a filtering in spatial direction with rounding and / or a filtering in channel direction with rounding, e.g. a lossless filtering, and / or a prediction processing in frequency direction with rounding and / or a prediction processing in time direction with rounding and / or a prediction processing in spatial direction with rounding and / or a prediction processing in channel direction with rounding, e.g. using unit 340, of the spectral domain representation 311 in order to obtain the filtered and / or prediction processed version of the spectral domain representation in the form of an integer

[0364] FH240905PEP-2025315045. DOCX representation and to encode the filtered and / or prediction processed version of the spectral domain representation, if the encoder is in the lossless version of the redundancy reduction operating mode.

[0365] Furthermore, the encoder may be configured to perform the filtering in frequency direction with rounding and / or the filtering in time direction with rounding and / or the filtering in spatial direction with rounding and / or the filtering in channel direction with rounding, e.g. a lossless filtering, and / or the prediction processing in frequency direction with rounding and / or the prediction processing in time direction with rounding and / or the prediction processing in spatial direction with rounding and / or the prediction processing in channel direction with rounding, of the precision reduced spectral domain representation 321 , in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation in the form of an integer representation and to encode the filtered and / or prediction processed precision reduced spectral domain representation, or at least a part thereof, if the encoder is in the near lossless version of the redundancy reduction operating mode.

[0366] Furthermore, the encoder may be configured to perform the filtering in frequency direction with rounding and / or the filtering in time direction with rounding and / or the filtering in spatial direction with rounding and / or the filtering in channel direction with rounding, e.g. a lossless filtering, and / or the prediction processing in frequency direction with rounding and / or the prediction processing in time direction with rounding and / or the prediction processing in spatial direction with rounding and / or the prediction processing in channel direction with rounding of the precision reduced spectral domain representation, in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation in the form of an integer representation and to encode a portion, e.g. as a true subset, of the filtered and / or prediction processed precision reduced spectral domain representation, e.g. to lossily encode the filtered precision reduced spectral domain representation, but no residual information, if the encoder is in the near lossless version of the noise shaping operating mode.

[0367] Furthermore, the encoder may be configured to perform a filtering in frequency direction without rounding and / or a filtering in time direction without rounding and / or a filtering in spatial direction without rounding and / or a filtering in channel direction without rounding, and / or a prediction processing in frequency direction without rounding, e.g. a lossy filtering, and / or a prediction processing in time direction without rounding and / or a prediction processing in spatial direction without rounding and / or a prediction processing in channel direction without rounding of the spectral domain representation 311 , in order to obtain a filtered and / or prediction processed version of the spectral domain representation and to encode a portion,

[0368] FH240905PEP-2025315045. DOCX e.g. as a true subset, of the filtered and / or prediction processed version of the spectral domain representation, if the encoder is in the lossy version of the noise shaping operating mode.

[0369] Moreover, the encoder 600 may, for example, be configured to encode an information 511 about the operating mode, and optionally an information about the respective version of the operating mode, of the encoder 600.

[0370] According to any of the above embodiments, the processing units (such as decoding / encoding units, filtering / prediction processing units, combining units, transform units, precision reduction units, residual determination units, encoding units and switching units) may be implemented by any suitable circuit structures, such as microprocessor circuits, ASIC circuits, CMOS circuits, and the like. According to embodiments, the processing units may be implemented as a combination of hardware structures and machine-readable instructions. For example, the processing units may include a processor and a storage device that stores machine-readable instructions that provide the described functionalities and lead to the execution of the methods described herein when executed by the processor. In examples, the storage device may be implemented by any suitable storage device, such as ROM, PROM, EPROM, EEPROM, flash memory, FRAM (ferroelectric RAM), MRAM (magnetoresistive RAM), or phase change RAM.

[0371] Reference is made to Fig. 1 to 6, to further highlight the inventive aspects of embodiments, which can be used individually or in combination.

[0372] Embodiments comprise encoders and decoders (e.g. apparatuses for decoding, e.g. apparatuses for encoding) enabling a coding of data, e.g. audio data, with scalable bitrate demands.

[0373] In particular, embodiments may enable a switching in between different operating modes, depending on the available bitrate, wherein a balance of prediction gain and noise shaping may be controlled selectively.

[0374] Hence, embodiments may comprise a separation of lossless signal codings chains and of lossy signal coding chains. A combination of said signal chains may enable operating modes in between, such as near lossless operating modes.

[0375] For a lossless processing, e.g. of the full spectral domain representation of the signal (e.g. in a fully lossless operating mode, e.g. given a sufficient bitrate), e.g. of the precision reduced spectral domain representation of the signal (e.g. as the losslessly coded portion of the spectral

[0376] FH240905PEP-2025315045. DOCX domain representation in a near lossless operating mode), the signal representations may be obtained as integer valued signal representations (e.g. IntMDCT (Integer Modified Discrete Cosine Transform) spectrums). In addition, the further processing, e.g. precision reduction, e.g. filtering / prediction processing may be performed using rounding operations, e.g. rounding operations which can be inverted in a respective decoder / encoder, to enable the lossless reconstruction.

[0377] For a lossy processing, e.g. of the full spectral domain representation of the signal (e.g. in a fully lossy operating mode, e.g. given a rather low bitrate), the signal representations may, for example, be obtained as non-integer valued signal representations (e.g. MDCT (Modified Discrete Cosine Transform) spectrums or even also as IntMDCT specrtums). In addition, the further processing, e.g. filtering / prediction processing may be performed without using rounding operations.

[0378] For example, the residual information of the signal (e.g. as the lossily coded portion of the spectral domain representation in a near lossless operating mode), a respective signal representation may be obtained as an integer valued signal representations (e.g. IntMDCT spectrums).

[0379] Hence, for example, embodiments may comprise the following operating modes:

[0380] A) Lossless mode

[0381] • Available bitrate is sufficient to transfer all bits of the original spectral domain representation 311 of the signal from encoder to decoder

[0382] • No precision reduction may be performed. Referring to Fig. 3 and 4: For example, the precision reduced spectral domain representation of the signal may be identical to the spectral domain representation of the signal and no residual information may be determined. Accordingly, signal 341 may be a filtered and / or prediction processed version of the original spectral domain representation. Referring to Fig. 5 and 6: For example, the spectral domain representation of the signal 311 may be provided to the filtering / prediction unit 340.

[0383] • The filtering / prediction processing unit (see 340) may operate with roundings, so that in the decoder, an inverse filtering / prediction processing, see 130, with a reconstruction of the rounding may be performed to obtain all bits of the original spectral domain representation (Note: In Fig. 1 and 2, the decoded version of the precision reduced spectral domain representation 111 may hence correspond to the encodersided filtered and / or prediction processed version of the original spectral domain

[0384] FH240905PEP-2025315045. DOCX representation and no encoded representation of a residual information 102 may be obtained).

[0385] • For example, the filtering / prediction processing unit (see 340) may be a TNS filter structure, which acts as a predictor (enabling a prediction gain), since no quantization noise is introduced, and the filtering is reversed in the decoder (e.g. inverse TNS “filtering” or respectively “prediction).

[0386] B) Near lossless mode

[0387] • Available bitrate is not sufficient to transfer all bits of the original spectral domain representation 311 of the signal from encoder to decoder. Hence, a certain amount of bits cannot be transferred

[0388] • From the original spectral domain representation 311 , the precision reduced spectral domain representation of the signal 321 and a residual information 331 may be obtained.

[0389] B1) (e.g. expected case or default case of B))

[0390] • The precision reduced spectral domain representation of the signal 321 may be obtained, so that the available bitrate is sufficient to losslessly encode the same or it is recognized that this is possible, e.g. given a predetermined determination of the precision reduced spectral domain representation of the signal 321 (e.g. a fixed number of bits).

[0391] • The filtering / prediction processing unit (see 340) may operate with roundings, so that in the decoder, an inverse filtering / prediction processing, see 130, with a reconstruction of the rounding may be performed to obtain all bits of the precision reduced spectral domain representation losslessly.

[0392] • For example, the filtering / prediction processing unit (see 340) may be a TNS filter structure, which only acts as a predictor (enabling a prediction gain) for the precision reduced spectral domain representation, since no quantization noise is introduced, and the filtering of the precision reduced spectral domain representation is reversed in the decoder (e.g. inverse TNS “filtering” or respectively “prediction) in unit 130.

[0393] • Depending on the available bitrate, some bits of the residual information may not be transmitted.

[0394] • However, a potential quantization noise included in the residual information, e.g. because some bits of the residual information could not be transmitted, is not filtered. Hence, even given strong predictions gains for the precision reduced representation, the quantization noise is not subject to such an undesirable influence of such “strong” filter parameters.

[0395] B2) (e.g. special case or fallback case of B))

[0396] FH240905PEP-2025315045. DOCX • One or more bits of the precision reduced spectral domain representation of the signal cannot be transferred from encoder to decoder.

[0397] • Nevertheless, the filtering / prediction processing unit (see 340) may operate with roundings.

[0398] • Decoder sided, a noise shaping at the inverse filtering / prediction processing unit 130 with roundings may be caused, however with limited power compared to a filtering of a truncated version of the full precision spectrum.

[0399] • As there is not even enough bitrate available for fully transmitting the precision reduced spectral domain representation, no residual information may be encoded (as the even less significant information)

[0400] • Decoder sided, the inversely filtered truncated precision reduced spectral domain representation of the signal may, optionally, be combined with a default (e.g. zero initialized) residual signal

[0401] C) Lossy mode

[0402] • Even a precision reduced spectral domain representation of the signal cannot be completely transmitted from encoder to decoder or a perceptually motivated noise shaping is desired or set

[0403] • Hence, no precision reduction may be performed. Referring to Fig. 3 and 4: For example, the precision reduced spectral domain representation of the signal may be identical to the spectral domain representation of the signal and no residual information may be determined (e.g. in Fig. 4 the case the signal 444 is provided to unit 350). Referring to Fig. 5 and 6: For example, the spectral domain representation of the signal 311 may be provided to the filtering / prediction processing unit 340.

[0404] • The filtering / prediction processing unit (see 340) may operate without roundings, so that in the decoder, using an inverse filtering / prediction processing, see 130, a perceptually noise shaped version of the spectral domain representation may be obtained.

[0405] • For example, the filtering / prediction processing unit (see 340) may be a TNS filter structure, which acts as a perceptual noise filter, since a significant amount of quantization noise is introduced and filtered in the decoder (e.g. inverse TNS filtering).

[0406] • Accordingly, the filterings (encoder-sided and decoder-sided) in this case may be performed without rounding

[0407] As a general remark, an encoder according to embodiments, e.g. encoder 300 or 400, may, for example, be configured to provide an encoded version of a filtered and / or prediction processed precision reduced spectral domain representation (e.g. in a near lossless operating

[0408] FH240905PEP-2025315045. DOCX mode) and / or for example, an encoded version of a filtered and / or prediction processed spectral domain representation (e.g. in a fully lossless or lossy operating mode) to a respective decoder.

[0409] Hence, when considering the full encoder-decoder processing chain, a respective encoded representation of a precision reduced spectral domain representation (e.g. 101) may be considered an encoded version of an encoder-sided filtered and / or prediction processed precision reduced spectral domain representation and an encoded representation of a spectral domain representation of a signal may be considered an encoded representation of an encoder-sided filtered and / or prediction processed spectral domain representation of a signal.

[0410] Accordingly, a respective decoder-sided filtering / prediction processing unit 130 may be configured to implement an inverted filtering / prediction compared to a respective encodersided filtering / prediction unit 330.

[0411] Accordingly, the decoder-sided filtered and / or prediction processed precision reduced spectral domain representation, e.g. 131 , may correspond to the encoder-sided precision reduced spectral domain representation 321 (e.g. in a case wherein the precision reduced spectral domain representation is transmitted and filtered and / or prediction processed in a lossless manner, so that the encoder-sided filtering / prediction processing and the decoder-sided filtering / prediction cancel each other out (e.g. at least approximately or substantially), e.g. using an invertible rounding).

[0412] Accordingly, a decoder-sided filtered and / or prediction processed spectral domain representation may correspond to the encoder-sided spectral domain representation 311 (e.g. in a case wherein the spectral domain representation is transmitted and filtered and / or prediction processed in a lossless manner, so that the encoder-sided filtering / prediction and the decoder-sided filtering / prediction cancel each other out (e.g. at least approximately or substantially)).

[0413] Hence, as a general remark regarding embodiments, the encoder may perform the inverse operation to “prediction” or respectively “filtering”. Regarding “prediction”, the encoder may be configured to remove the predictable parts of the signal, which the decoder then predicts again, e.g. in the sense of a reconstruction.

[0414] Hence, as a general aspect of embodiments, the encoder may be configured to generate a kind of excitation representation from the original spectral signal representation. This may, for

[0415] FH240905PEP-2025315045. DOCX example, be done in such a way that when the decoder-sided prediction processing structure processes the excitation representation as input, it predicts, i.e., reconstructs, the original spectral signal representation.

[0416] To put in other words, the encoder may, for example, perform the inverse operation to “prediction” (e.g. decoder-sided prediction, e.g. decoder-sided prediction processing). The encoder may hence remove the predictable parts of the respective signal, which the decoder then predicts again, i.e., reconstructs. Hence, as a general example for embodiments, the encoder may, for example, generate a kind of excitation representation from the original spectral signal representation (or a precision reduced version thereof). This may, for example, be done in such a way that when the decoder-sided prediction processing receives the excitation representation (e.g. a decoded version thereof) as input, it may predict, i.e., reconstruct, the original spectral signal representation.

[0417] Hence, a difference between the decoder-sided “prediction” and the encoder-side inverse operation for “prediction” may be considered a redundancy reduction, encoder-sided, and a redundancy reconstruction, decoder sided.

[0418] Hence, the encoder sided prediction processing may be considered an “inverse prediction”, as opposed to a prediction, e.g. in the form of a “forward prediction” in the decoder).

[0419] In the following, further different inventive embodiments and aspects will be described. Also, previously discussed embodiments will be explained in different words, or additional, optional features, functionalities and details of previously discussed embodiments will be disclosed.

[0420] It should be noted that any embodiments disclosed herein can be supplemented by any of the details (features and functionalities) described in the following description.

[0421] Also, the embodiments described in any of the following paragraphs can be used individually, and can also be supplemented by any of the features in another paragraphs, or by any feature included in the claims.

[0422] Also, it should be noted that individual aspects described herein can be used individually or in combination. Thus, details can be added to each of said individual aspects without adding details to another one of said aspects.

[0423] FH240905PEP-2025315045. DOCX It should also be noted that the present disclosure describes, explicitly or implicitly, features usable in an audio encoder (apparatus for providing an encoded representation of an input audio signal) and in an audio decoder (apparatus for providing a decoded representation of an audio signal on the basis of an encoded representation). Thus, any of the features described herein can be used in the context of an audio encoder and in the context of an audio decoder.

[0424] Moreover, features and functionalities disclosed herein relating to a method can also be used in an apparatus (configured to perform such functionality). Furthermore, any features and functionalities disclosed herein with respect to an apparatus can also be used in a corresponding method. In other words, the methods disclosed herein can be supplemented by any of the features and functionalities described with respect to the apparatuses.

[0425] Moreover, features and functionalities disclosed herein relating to a method or apparatus can also be used in a bistream (having a structure as defined by the method and or processing performed by the apparatus). In other words, an inventive bitstream can be supplemented by any of the features and functionalities described with respect to an apparatuses and / or method.

[0426] Also, any of the features and functionalities described herein can be implemented in hardware or in software, or using a combination of hardware and software, as will be described in the section “Implementation Alternatives”.

[0427] The following section may be titled TNS for (near) lossless audio coding:

[0428] First, reference is made to a short overview of conventional approaches. In this regard, it is to be noted that embodiments may comprise some or even all of the features discussed and / or disclosed in the following, individually or taken in combination.

[0429] Temporal Noise Shaping is a well-known coding tool and for instance used in MPEG-4 AAC- LC or ETSI LC3plus. The basic concept is described in [1] and [2],

[0430] Basically, a prediction filter in frequency domain on the encoder side may remove energy from the signal which may, at least partially, be restored again on the decoder side by the inverse filter (e.g. in general by filtering). Any quantization error added in between may be shaped temporally by the decoder side inverse filter (e.g. in general by filtering). All TNS implementations so far focus on the noise shaping aspect.

[0431] FH240905PEP-2025315045. DOCX In [2] it is outlined, that the (temporal) noise shaping effect of TNS is achieved by applying the frequency domain filter as an open-loop prediction scheme, and [2] also mentions the alternative of applying the frequency domain filter as a closed-loop prediction scheme, which results in a temporally flat quantization error. For perceptual audio coding it is concluded, that the open-loop prediction scheme is preferred due to the temporal shaping of the quantization error.

[0432] In [3] a scalable perceptual and lossless audio coding scheme based on MPEG-4 AAC is presented. In the lossless enhancement layer TNS is applied on the IntMDCT spectrum as a lossless prediction scheme. In detail, Section 5.3.2 of [3] states:

[0433] "The TNS tool in AAC modifies the MDCT spectrum by applying linear prediction filters before applying quantization. Consequently the difference between the IntMDCT values of the lossless enhancement layer and the quantized MDCT values of the AAC core layer would increase. The TNS tool can, however, also be applied to the IntMDCT values in a lossless way by using the same prediction filter and including a rounding to integer values after each prediction step. In the decoder, the original IntMDCT spectrum is reconstructed by using the inverse filter and the same rounding."

[0434] Basically, this lossless prediction using TNS is similar to the lossless predictive coding in time domain used in many time domain based lossless audio coding schemes, as outlined in [5] with the following Fig. 7. Fig. 7 shows schematic views of different coding structures on which embodiments may be based.

[0435] The upper version (encoder left, decoder right) describes the floating version used for TNS as noise shaping tool, where the quantization of the spectrum (see e.g. Q) is applied between the filter at encoder side and its inverse one. The lower figure describes the integer based version / lossless version where each filter step is rounded / truncated in order to maintain the precision of the data. The main difference is that the lossless prediction using TNS works in spectral domain and not in time domain.

[0436] In [4] the same structure as in [3] is used for building the scalable perceptual and lossless audio coding scheme HD-AAC based on MPEG-4 AAC, which is standardized as MPEG-4 SLS in ISO / IEC 14496-3. In addition to the structure in [3], HD-AAC also provides a nearlossless mode based on bit-plane coding in the lossless enhancement layer. If not all bit-planes are decoded by the decoder, the codec operates in a near-lossless mode by quantizing the IntMDCT spectrum with less accuracy than that required for lossless reconstruction. The

[0437] FH240905PEP-2025315045. DOCX inverse TNS filter in the HD-AAC decoder operates on the transmitted IntMDCT, regardless of whether this was transmitted completely for lossless decoding or quantized for near-lossless decoding. Thus, for near-lossless decoding the TNS filter turns into a classical TNS prediction scheme with noise shaping for near-lossless coding, while it is a lossless prediction scheme for lossless coding.

[0438] Regarding embodiments, It is to be noted that “closed loop” may, for example be understood, or may, for example mean, that the prediction can be completely (or, can be, for example, at least substantially) inverted, for example as all information is available. If this is not the case, the term “open loop” may, for example, be used. These terms may, on the one hand, optionally be understood as defined in the literature, but are not limited to such definitions and may hence, on the other hand, as well be understood in a slightly different or different manner, as will be apparent for the person skilled in the art in view of the present disclosure.

[0439] Hence, optionally, embodiments may comprise features, functionalities and details, regarding conventional open loop and / or closed loop filtering approaches, in particular TNS filtering approaches (e.g. well known features, which will not be explained or mentioned in detail, in order not to obscure the features of embodiments of the invention).

[0440] According to some embodiments, “open-loop”, or “or open-loop TNS, may be understood as “lossy TNS”, or “lossy encoding approach” and “closed loop” may be understood as “lossless TNS”, or “lossless encoding approach”. In particular, embodiments may implement variations or improved versions of “open-loop TNS” (e.g. in a sense of an open-loop prediction gain filtering) and / or “closed-loop TNS” (e.g. in a sense of a closed-loop prediction gain filtering), e.g. using a same filter structure or architecture.

[0441] Regarding optimized near lossless operation according to embodiments, the following is to be noted:

[0442] Adaptive lossless codecs can, for example, switch between a lossless and a perceptual optimal lossy mode depending on the available bitrate and input signal. If enough bits are available, the current frame may, for example, be coded in a lossless way. As TNS is, for example, a substantial, or important, or for example even a required tool for the perceptual coding mode, TNS can also work without noise shaping operation in the lossless mode. TNS may, for example, therefore be switched between a noise shaping mode and a lossless prediction mode.

[0443] FH240905PEP-2025315045. DOCX The noise shaping operation may cause a significant change of the signal’s amplitude, for example, esp. if the TNS prediction gain and the quantization error are high. Perceptually this behavior is envisioned, however for near lossless codecs, preserving most of the MSBs is preferred.

[0444] Fig. 8 shows an example of how the MSE (Mean Squared Error) of the spectral data is increased by the active TNS filter for some dedicated frames. This may happen if not all LSBs of the spectral lines can be transferred from the encoder to the decoder side which may, for example, correspond to a quantization error. Fig. 8 shows the Mean Square Error, MSE, over the number of LSBs removed, for different frames (frame 3, frame 34 and frame 332). The top of Fig. 8 shows the MSE for input of TNS decoder (d-TNS-dec), the bottom of Fig. 8 shows the MSE for an output of the TNS decoder (d-dec). Hence, it was recognized that although only LSBs are removed, strong prediction gains of TNS may amplify an influence of such quantization noise, so as to have a significant impact on the MSE.

[0445] To minimize this effect, the dynamic range for active TNS frames may, for example, be reduced to enable, or for example even to ensure that all (or for example substantially all) bits are transferred to decoder side. Therefore, whenever the TNS filter shows a significant prediction gain and the filter is active, the MDCT spectrum (e.g. in general the spectral domain representation) may, for example be split into a spectrum with reduced precision and it’s residual (in general embodiments do not rely on a split of the spectrum but, according to some embodiments based on the MDCT spectrum a precision reduced spectral domain representation and a residual information may be obtained). The optional split may be implemented as MSB / LSB split where MSBs and LSBs are separated at a certain bit plane.

[0446] The TNS filter may, for example, operate on the spectrum with reduced precision and the prediction filter may, for example, run as a lossless prediction system. The reduced precision spectrum may, for example, require less bits to code and therefore, a lossless coding chain at a certain target bitrate can be enabled between TNS filter and the inverse TNS filter.

[0447] The residual spectrum may, for example, be coded separately by a residual coder and optionally be added to the bitstream. For example, depending on the final data rate, some of the residual bits might be removed. The whole process may, for example, be inverted at the decoder side.

[0448] This scheme allows that most of the prediction capability of the TNS module is maintained while the amplification of the quantization error is not present (or at least reduced). This allows

[0449] FH240905PEP-2025315045. DOCX a TNS module design which can operate from low bitrate with high noise shaping capability to higher rates with the focus on prediction gain only (e.g. using a noise shaping operating mode and a redundancy reduction operating mode).

[0450] Fig. 9 outlines an example for the processing at encoder and decoder side (all details are optional). Fig. 9 shows a schematic view of a coding concept according to embodiments. Hence, an encoder may be provided with an Integer MDCT spectrum 911 (e.g. corresponding to the spectral domain representation 311 of the signal). The encoder may be configured to determine whether a filtering, e.g. a temporal noise shaping (as shown in Fig. 9) is to be used, see 910. Optionally, as an example, e.g. according to some embodiments, the encoder may, for example, be configured to determine whether a prediction processing is to be used.

[0451] For example, in case such a filtering is not active, the original spectrum may be provided to coding unit 950 (e.g. a scalable lossless to lossy coder). Unit 950 may represent a respective encoding unit and decoding unit, simplified to a single block for the sake of brevity in Fig. 9.

[0452] For example, in case such a filtering is active, the spectrum 911 may be split, e.g. using a spectrum splitter 920 (e.g. Spektrum spliiter), e.g. corresponding to precision reduction unit 320 and residual determination unit 330. In other words, based on the spectral domain representation 911 , a precision reduced spectral domain representation (e.g. a precision reduced spectrum) 921 (e.g. corresponding to 321) and a residual information 931 (e.g. a residual of the reduced spectrum, e.g. as a difference between 911 and 921) may be obtained. Hence, in general the spectrum splitter 920 may, for example, be considered a combination of precision reduction unit 320 and residual determination unit 330.

[0453] The precision reduced spectral domain representation 921 may subsequently be filtered and / or prediction processed, e.g. using a TNS filter 940 (e.g. corresponding to 340), in order to obtain a filtered and / or prediction processed precision reduced spectral domain representation 941 (e.g. corresponding to 341).

[0454] The residual information 931 may, for example be coded by a distinct coder 960, to obtain a representation 961 of the coded residual information, e.g. residual bits.

[0455] Hence with regard to Fig. 3 and 4 it is to be noted that encoders 300, 400 may comprise distinct encoding units for the filtered and / or prediction processed precision reduced spectral domain representation and for the residual information.

[0456] FH240905PEP-2025315045. DOCX As shown in Fig. 9 (bottom half), decoder sided, the coding chain may be applied in reverse. In detail, the decoder may be provided with an information or may determine whether a respective filtering was active, see 910’, for the encoded data (e.g. using a side information, e.g. using an operating mode information). If this is not the case, the encoded representation of the spectral domain representation of the signal may be decoded, in order to obtain the decoded Integer MDCT spectrum 91T.

[0457] If such a filtering was applied encoder-sided, based on the bitstream, e.g. as distinct bitstream portions, or based on a common encoded representation, a decoded version of a precision reduced spectral domain representation 94T (e.g. corresponding to 111) and a decoded version of the residual information 961’ (e.g. corresponding to 121 , e.g. using an inverse residual coder 960’) is obtained.

[0458] The decoded version of the precision reduced spectral domain representation 94 T is then provided to filtering / prediction processing unit 940’ (e.g. corresponding to 130), e.g. providing an inverse filtering / prediction processing functionality, compared to unit 940, in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation 92 T (e.g. corresponding to 131).

[0459] Using a spectrum combiner 920’, (e.g. corresponding to unit 140, e.g. a Spektrum combiner) based on the filtered and / or prediction processed precision reduced spectral domain representation 92T and the decoded version of the residual information 96T a reconstructed version 91T of the Integer MDCT spectrum may be obtained.

[0460] In the following, further examples for operation cases when TNS is active, see e.g. 910, 910’, are discussed (wherein, for example, it is not necessary that all operation cases are implemented in an embodiment):

[0461] A) Lossless mode (optional)

[0462] • All bits can be transferred from encoder to decoder

[0463] • Spectrum splitter, e.g. 920, moves all bits to MSB (precision is not reduced, for example, 921 may correspond to 911); No residual signal, e.g. 931 , available

[0464] • TNS filter (e.g. in the encoder) operates with roundings on complete spectrum

[0465] • Internal coder is lossless (e.g. at the side of the encoder, and possibly also on the side of the decoder)

[0466] • No noise shaping at inverse TNS filter (e.g. in the decoder, see e.g. 940’) with rounding

[0467] FH240905PEP-2025315045. DOCX B) Near lossless mode

[0468] • For example (e.g. in some situations), a certain amount of bits cannot be transferred from encoder to decoder

[0469] • Spectrum splitter (e.g. in the encoder, e.g. 920) generates spectrum with reduced precision, e.g. 921 , and residual, e.g. 931

[0470] • TNS filter (e.g. in the encoder, e.g. 940) with roundings applied on spectrum with reduced precision, e.g. 921

[0471] • B1) (Expected case) o All bits of reduced precision spectrum can be transferred from encoder to decoder o Internal coder (e.g. in the encoder and possibly also in the decoder) is lossless o No noise shaping at inverse TNS filter (e.g. in the decoder, e.g. 940’) with roundings of spectrum with reduced precision o For example, all or parts of residual coder, e.g. 960’, generate residual signal o Spectrum combiner (e.g. in the decoder, e.g. 920’) combines lossless spectrum of reduced precision, e.g. 92T and residual signal, e.g. 96T o For example, potential error signal due to missing residual signal is not shaped or amplified by TNS filter with roundings, e.g. 940’

[0472] • B2) (Fallback strategy) (Optional) o Some few bits of reduced precision spectrum, e.g. 921 , cannot be transferred from encoder to decoder o For example, internal coder, e.g. 950, truncates the bits (lossy) o Noise shaping at inverse TNS filter (e.g. in the decoder, e.g. 940’) with roundings of spectrum with reduced precision happens, however, for example, with very limited power compared to full precision spectrum o For example, residual coder (e.g. in the decoder, e.g. 960’ and / or e.g. in the encoder 960 (e.g. to not encode a residual signal)) skips residual signal o For example, spectrum combiner (e.g. in the decoder, e.g. 920’) combines spectrum of reduced precision e.g. 92T and zero initialized residual signal (e.g. instead of residual of reduced spectrum 96T) o For example, error signal due to missing residual signal is not shaped or amplified by TNS filter with roundings

[0473] C) Lossy mode (optional)

[0474] FH240905PEP-2025315045. DOCX • Even spectrum with reduced precision, e.g. 921 , cannot completely be transmitted from encoder to decoder or noise shaping perceptually motivated

[0475] • Spectrum splitter (e.g. in the encoder, e.g. 920) moves all bits to MSB (precision is not reduced); No residual signal, e.g. 931 , available

[0476] • TNS filter (e.g. in the encoder e.g. 940) without roundings works on complete spectrum

[0477] • Internal coder (e.g. in the encoder and possibly in the decoder) is perceptually lossy

[0478] • Noise shaping at inverse TNS filter (e.g. in the decoder e.g. 940) active without roundings

[0479] It is to be noted that for the lossless or near lossless operation a rounding for the filter, e.g. 940, e.g. 940’, may be implemented, which may, for example even be required. Hence, in the above figure, the encoder-sided TNS filter and / or the decoder-sided inverse TNS filter may optionally comprise rounding functionalities, e.g. to provide integer outputs.

[0480] Fig. 10 shows a schematic view of an example of the data flow (e.g. data float) in all operation modes according to an embodiment (all details optional, in particular less modes may be present than shown). Fig. 10 shows decoder operation cases (e.g. operating modes), which may, for example, be switched in dependence on an available bitrate and / or an operating mode information (e.g. as a part of a side information), e.g. indicating a noise shaping operating mode, e.g. a perceptual operating mode or e.g. indicating a redundance reduction operating mode.

[0481] From a respective encoder, a decoder according to embodiments may receive an encoded version of a filtered and / or prediction processed spectral domain representation of a signal, see as an example the TNS filtered integer MDCT spectrum 1001.

[0482] For example, based on an operating mode information or based on an information about an available bitrate, the decoder may switch between a filtering / prediction processing of the spectrum 1001 with rounding (e.g. w / rounding), see 1010 or without rounding (e.g. w / o rounding), see 1020.

[0483] As discussed before, in the context of the full coding chain, the decoder-sided filtering / prediction processing may be considered an inverse filtering / prediction processing with respect to the encoder-sided filtering, hence here as an example an inverse TNS filtering.

[0484] FH240905PEP-2025315045. DOCX Depending on the operating mode or on the available bitrate, a decoder according to embodiments may be provided with residual data, see 1002 (e.g. an encoded version thereof) or not. If residual data 1002 is received, the same may be decoded, using residual coder 1030 (which may be again considered an inverse coder with respect to a corresponding residual coder of an encoder).

[0485] If no residual data is provided, default values, such as zeros, may be used as residuals data, see 1003, or an incorporation of residual data may be fully disabled, see 1004.

[0486] Depending on the operating mode, the spectrum combiner 1020 (e.g. spektrum combiner) may combine inversely filtered spectral information (top signal path in Fig. 10) and residual information, (bottom signal path in Fig. 10) or may provide the MDCT spectrum 1021 only based on inversely filtered spectral information.

[0487] In particular, the above-discussed operating modes A, B, with B1 and B2 and C may be implemented as illustrated with Fig. 10. A respective encoder may be configured to perform a corresponding switching of signal coding chains and operating modes.

[0488] According to embodiments, rounding operations may, optionally, be performed or may happen at the following entities, or in other words, the following list may indicate where rounding operations optionally happen

[0489] Internal decoder o Lossy mode - Quantization, e.g. Quantization of spectral data o Lossless mode, if residual coder truncates information

[0490] Filtering and / or prediction processing (e.g. TNS) in lossless or near lossless mode o Residual coder, if residual coder truncates information o For example, for each filtering step, rounding operation may be performed or for example even be required; optionally implicitly done for fixed-point arithmetic, e.g. if signal is scaled to 16 bit or 24 bit dynamic

[0491] In the above-description, the Integer MDCT spectrum may be considered as an example for a spectral domain representation of a signal. The spectrum splitter may be considered as an example implementation for obtaining the precision reduced spectral domain representation, here, as an example, in the form of the precision reduced spectrum, and the residual information, here, as an example, in the form of the residual of reduced spectrum. Furthermore, the TNS filter may be considered as an example for a filtering in a frequency direction.

[0492] FH240905PEP-2025315045. DOCX Embodiments according to the invention comprise

[0493] Encoder to produce spectrum of reduced precision and apply TNS filter in lossless operation mode on reduced precision spectrum o Special case of reduced precision is MSB / LSB split o Encode filtered spectrum or part of o Create and encode residual spectrum or part of Decoder is inverse

[0494] In the following, results according to embodiments are discussed. Reference is made to Fig. 11 : This schematic scatter plot in Fig. 11 illustrates the segmental SNR values of all frames where Temporal Noise Shaping (TNS) is active, comparing a TNS version with adaptive spectrum splitter, see 1110, to generate a precision reduced spectrum to a TNS version without the spectrum splitter, see 1120, according to an embodiment. Each dot represents a frame's SNR value derived from a dataset of transient audio items (48 kHz, 16-bit) encoded using LC3plus in lossless mode at a bitrate of 400 kbps (Fig. 11 shows SNR per frame [dB] for TNS active frames in transient items at 400kbit 48kHz 16bit) .

[0495] Conventional metrics like mean or median do not account for SNR values of "Inf," which indicate lossless frames. Therefore, this scatter plot effectively showcases the improvements offered by the new method (in general methods according to embodiments) by highlighting the percentage of frames that exhibit enhanced performance (depicted in blue, see 1101 (Lossy TNS Frames), on the left side of the red diagonal line 1102 (e.g. y=x)) or achieve lossless quality (shown in green, see 1103 (Lossless TNS Frames), on the left side). In this case, "Inf" is mapped to 100 dB.

[0496] The data points along the diagonal (red line, see 1102) represent frames where both methods yield identical SNR values (Equal: 31.62%, see red line 1102). Notably, 10.37% of frames are worse with the new method, however these instances show insignificant numerical differences in terms of SNR compared to the standard TNS method. In contrast, the improvements on the left side of the diagonal are substantial, with 41.29% of frames demonstrating enhanced SNR values and 16.72% achieving lossless quality.

[0497] Hence, in other words, Fig. 11 (e.g. showing SNR PER frame) shows segmental SNR of original and decoded audio signal comparing TNS configurations with and with (e.g. or without) spectrum splitter and precision reduced spectrum.

[0498] FH240905PEP-2025315045. DOCX In the following, in view of the above-disclosure, some features of embodiments will be summarized. The feature mentioned here may be used individually, and may optionally be introduced into any of the embodiments disclosed herein:

[0499] According to embodiments of the invention, an architecture may be implemented comprising an encoder, which is configured to obtain a precision reduced spectral domain representation of a signal, such as an audio signal, based on a spectral domain representation of the signal (e.g. as a result of an Integer MDCT transform). Optionally, a residual information may be obtained. The residual may describe a difference or an approximation of (e.g. between) a difference of the spectral domain representation of the signal and the precision reduced version thereof.

[0500] Furthermore, for a subsequent transmission from the encoder to a respective decoder, the energy of the precision reduced spectral domain representation may be reduced by a filtering, so as to obtain a filtered precision reduced spectral domain representation.

[0501] An optional transmission of the residual information may be implemented according to embodiments, for example, by encoding a residual spectral domain representation or a portion thereof (as an example for the residual information). Optionally, an information about the residual information may as well be provided in the form of indices to default or predetermined residual spectral domain representations (e.g. spectral coefficients), for example, for use in a look up table in a respective decoder. However, the optional information about the residual information may as well be absent in a respective bitstream, e.g. so as to be interpreted by a respective decoder to use default values for the residual information. The amount of residual information provided may be scaled according to an available bitrate.

[0502] The filtering of the precision reduced spectral domain representation may, in particular, be achieved using a prediction filter, e.g. a temporal noise shaping filter.

[0503] In a respective decoder, the precision reduced spectral domain representation may be obtained and filtered, e.g. using a corresponding (corresponding to the encoder-side) prediction filter, e.g. a temporal noise shaping filter.

[0504] The filtering in the decoder may be implemented as an inverse filtering to the encoder-sided filtering, in order to restore at least a portion of the original signal energy of the precision reduced spectral domain representation.

[0505] FH240905PEP-2025315045. DOCX Hence, it is to be noted that a filtered precision reduced spectral domain representation in the decoder may correspond (in view of the system comprising encoder and decoder) to an inversely filtered - filtered precision reduced spectral domain representation, which may hence correspond, at least approximately, to the precision reduced spectral domain representation (so that the filtering in the encoder and the inverse filtering in the decoder cancel each other out, e.g. at least substantially).

[0506] Accordingly, the filter structures and / or filter parameters used in encoder and decoder may be set, so as to achieve a temporal noise shaping, if noise, e.g. quantization noise, is introduced between the filtering in the encoder and the (inverse) filtering in the decoder to the filtered precision reduced spectral domain representation.

[0507] Hence, regarding embodiments, a temporal noise shaping filter structure and / or temporal noise shaping filter parameters may be understood as a structure and / or parameters for use in a respective encoder-sided filter and / or in a respective decoder-sided (inverse) filter, so as to achieve a temporal noise shaping, if noise, e.g. quantization noise, is introduced between the filtering in the encoder and the (inverse) filtering in the decoder to the filtered precision reduced spectral domain representation.

[0508] First, it was recognized that obtaining the precision reduced spectral domain representation, based on the spectral domain representation, allows enabling or even ensuring a lossless coding or at least a near lossless coding of the precision reduced portion.

[0509] Second, it was recognized that a same filtering structure and / or same filtering parameters (e.g. depending on a respective input signal, the parameters hence being determined signal dependent, but not operating mode (e.g. TNS or prediction gain) dependent) may be used on the one hand to achieve a temporal noise shaping and on the other hand to achieve a prediction gain. In other words, embodiments may comprise corresponding (e.g. filter and inverse filter) structures in encoder and decoder and, for example, depending on the specific characteristics of the signal to be transmitted, from encoder to decoder, a set of filter parameters (e.g. for filter and inverse filter) may be determined. Now, it was recognized that using the same architecture and optionally parameters two different operating modes may be implemented, namely a temporal noise shaping operating mode and a redundancy reduction operating mode (e.g. a prediction gain operating mode). The difference in operation may be achieved by allowing or not allowing an introduction of quantization noise in the precision reduced spectral domain representation, wherein the allowing or not allowing the introduction of noise is achieved by the precision reduction of the spectral domain representation. Hence,

[0510] FH240905PEP-2025315045. DOCX the filter parameters may be determined based on characteristics of the signal to be encoded but agnostic of the operating mode, e.g. irrespective of whether a prediction gain or a temporal noise shaping is to be achieve, e.g. irrespective of a lossy or a lossless encoding is to be performed (e.g. for the precision reduced spectral domain representation).

[0511] Hence, obtaining the precision reduced spectral domain representation allows selectively activating and scaling a temporal noise shaping and prediction gain functionality (e.g. a combination thereof with scalable impacts of the two functionalities on the processed signal that is encoded and decoded), for example, depending on an available bitrate.

[0512] In other words, the precision reduction, e.g. based on a splitting of the spectral domain representation into the precision reduced spectral domain representation and a residual information (e.g. as an example in the form of MSBs and LSBs), allows enabling or even ensuring that the precision reduced portion can be transmitted losslessly, in order to achieve a prediction gain and optionally to prevent temporal noise shaping on this portion.

[0513] In particular, as outlined above, it was recognized that such a prediction gain focused functionality may be achieved by using a temporal noise shaping filter structure or even a temporal noise shaping filtering (e.g. using a filter having a filter structure and parametrization that would allow to effect a temporal noise shaping, if a quantization error is introduced to the respective signal, e.g. after filtering in a respective encoder and before applying a corresponding inverse filtering in a respective decoder).

[0514] On the other hand, with the same architecture, comprising a respective same filter structure (e.g. temporal noise shaping filter structure, e.g. with corresponding filters and respectively inverse filters in encoder and decoder), a seamless transition (e.g. from a prediction gain functionality) to a temporal noise shaping functionality can be achieved by allowing an introduction of quantization noise between the filtering in the encoder and a respective inverse filtering in the decoder.

[0515] Nevertheless, the filter parameters may be provided signal and / or time dependent, e.g. framewise, from encoder to decoder, but for example irrespective of whether quantization noise is introduced to the signal or not.

[0516] The approach comprises a good scalability, based on the precision reduction of the spectrum. For example, the precision reduction can be performed to allow an introduction of a certain

[0517] FH240905PEP-2025315045. DOCX amount of quantization noise, in order to increase an amount, or even introduce an amount (at all) of a temporal noise shaping, effected in the decoder sided (inverse) filter.

[0518] Hence, different operating modes can be obtained:

[0519] • Lossless Mode, e.g. wherein the whole spectral domain representation of the signal is provided as the precision reduced spectral domain representation of the signal and hence filtered, transmitted, inversely filtered and fully reconstructed (because of an available bitrate enabling lossless transmission of the whole spectrum). In this case, for example, no residual information may be determined and transmitted.

[0520] • Near lossless mode, wherein the precision reduction is performed, so as to enable a lossless encoding and transmission of the precision reduced spectral domain representation of the signal and wherein portions of the residual information can be scalably transmitted or omitted. For example, if an available bitrate drops, the residual information can be omitted completely and even some portions, e.g. some bits, of the precision reduced spectral domain representation can be lost to quantization, so that in the decoder a limited (e.g. controllable and limited) temporal noise shaping is introduced.

[0521] • Lossy Mode, e.g. wherein the whole spectral domain representation of the signal is provided as the precision reduced spectral domain representation of the signal and hence filtered, transmitted, inversely filtered and lossily reconstructed (because of an available bitrate only enabling a lossy transmission of the precision reduced spectral domain representation). In this case, for example, no residual information may be determined and transmitted.

[0522] According to embodiments, in order to achieve a fully reconstructable filtering - inverse filtering step, optionally a rounding may be implemented. Hence, a processing, for example, in integer domain may be enabled. The rounding may for example be used in the lossless and near lossless modes, but not in the lossy case.

[0523] Hence, with regard to the different operating modes, a decoder (as well as an encoder) according to embodiments may comprise a mode with activated rounding for the filtering and with deactivated rounding for the filtering.

[0524] Furthermore, it is to be noted that embodiments comprise approaches or are related to TNS for near lossless audio coding. Embodiments may address or comprise lossless audio coding,

[0525] FH240905PEP-2025315045. DOCX e.g. near lossless audio coding, and / or an optimal selection of noise shaping and prediction capability.

[0526] Embodiments may be used in particular with regard to the field of lossless and low delay audiocoding and may hence comprise and / or enable such functionalities. Furthermore, embodiments may be implemented in LC3plus lossless or AAC-ELD lossless.

[0527] Temporal Noise Shaping is a well-known coding tool and for instance used in MPEG-4 AAC- LC or ETSI LC3plus. Adaptive lossless codecs according to embodiments can switch between a lossless and a perceptual optimal lossy mode, e.g. depending on the available bitrate and input signal. As TNS is an important or, for example, even required tool for the perceptual coding mode, it was recognized that TNS can also work without noise shaping operation in the lossless mode. According to embodiments, TNS may therefore be switched between a noise shaping mode and a prediction only mode (e.g. a redundancy reduction operating mode). For near lossless, TNS may operate on the signal or a signal with reduced precision, which may require less bits to code. Therefore, a lossless coding path can be enabled between TNS coder and decoder where TNS may, for example, only operate as predictor.

[0528] Implementation Alternatives

[0529] Although some aspects have been described in the context of an apparatus, it is clear that these aspects also represent a description of the corresponding method, where a block or device corresponds to a method step or a feature of a method step. Analogously, aspects described in the context of a method step also represent a description of a corresponding block or item or feature of a corresponding apparatus. Some or all of the method steps may be executed by (or using) a hardware apparatus, like for example, a microprocessor, a programmable computer or an electronic circuit. In some embodiments, one or more of the most important method steps may be executed by such an apparatus.

[0530] Depending on certain implementation requirements, embodiments of the invention can be implemented in hardware or in software. The implementation can be performed using a digital storage medium, for example a floppy disk, a DVD, a Blu-Ray, a CD, a ROM, a PROM, an EPROM, an EEPROM or a FLASH memory, having electronically readable control signals stored thereon, which cooperate (or are capable of cooperating) with a programmable computer system such that the respective method is performed. Therefore, the digital storage medium may be computer readable.

[0531] FH240905PEP-2025315045. DOCX Some embodiments according to the invention comprise a data carrier having electronically readable control signals, which are capable of cooperating with a programmable computer system, such that one of the methods described herein is performed.

[0532] Generally, embodiments of the present invention can be implemented as a computer program product with a program code, the program code being operative for performing one of the methods when the computer program product runs on a computer. The program code may for example be stored on a machine-readable carrier.

[0533] Other embodiments comprise the computer program for performing one of the methods described herein, stored on a machine-readable carrier.

[0534] In other words, an embodiment of the inventive method is, therefore, a computer program having a program code for performing one of the methods described herein, when the computer program runs on a computer.

[0535] A further embodiment of the inventive methods is, therefore, a data carrier (or a digital storage medium, or a computer-readable medium) comprising, recorded thereon, the computer program for performing one of the methods described herein. The data carrier, the digital storage medium or the recorded medium are typically tangible and / or non-transitionary.

[0536] A further embodiment of the inventive method is, therefore, a data stream or a sequence of signals representing the computer program for performing one of the methods described herein. The data stream or the sequence of signals may, for example, be configured to be transferred via a data communication connection, for example via the Internet.

[0537] A further embodiment comprises a processing means, for example a computer, or a programmable logic device, configured to or adapted to perform one of the methods described herein.

[0538] A further embodiment comprises a computer having installed thereon the computer program for performing one of the methods described herein.

[0539] A further embodiment according to the invention comprises an apparatus or a system configured to transfer (for example, electronically or optically) a computer program for performing one of the methods described herein to a receiver. The receiver may, for example,

[0540] FH240905PEP-2025315045. DOCX be a computer, a mobile device, a memory device or the like. The apparatus or system may, for example, comprise a file server for transferring the computer program to the receiver.

[0541] In some embodiments, a programmable logic device (for example a field programmable gate array) may be used to perform some or all of the functionalities of the methods described herein. In some embodiments, a field programmable gate array may cooperate with a microprocessor in order to perform one of the methods described herein. Generally, the methods are preferably performed by any hardware apparatus.

[0542] The apparatus described herein may be implemented using a hardware apparatus, or using a computer, or using a combination of a hardware apparatus and a computer.

[0543] The apparatus described herein, or any components of the apparatus described herein, may be implemented at least partially in hardware and / or in software.

[0544] The methods described herein may be performed using a hardware apparatus, or using a computer, or using a combination of a hardware apparatus and a computer.

[0545] The methods described herein, or any components of the apparatus described herein, may be performed at least partially by hardware and / or by software.

[0546] The above-described embodiments are merely illustrative for the principles of the present invention. It is understood that modifications and variations of the arrangements and the details described herein will be apparent to others skilled in the art. It is the intent, therefore, to be limited only by the scope of the impending patent claims and not by the specific details presented by way of description and explanation of the embodiments herein.

[0547] FH240905PEP-2025315045. DOCX References

[0548] [1] Jurgen Herre: TEMPORAL NOISE, SHAPING, QUANTIZATION AND CODING METHODS IN PERCEPTUAL AUDIO CODING: A TUTORIAL INTRODUCTION, 1999, Paper 17-031

[0549] [2] J. Herre and J. D. Johnston, “Enhancing the Performance of Perceptual Audio Coders by Using Temporal Noise Shaping (TNS),” in 101st AES Convention, Los Angeles, 1996, preprint 4384.

[0550] [3] Ralf Geiger, Scalable Perceptual and Lossless Audio Coding based on MPEG-4 AAC, 115thAES Convention, 2003, Paper Nr. 5868

[0551] [4] Ralf Geiger, Rongshan Yu, et al., "ISO / IEC MPEG-4 High-Definition Scalable Advanced Audio Coding ," in 120th AES Convention, Paris, France, 2006, preprint 6791.

[0552] [5] https: / / www.eetimes.com / lossless-compression-of-audio-data-part-2 /

[0553] FH240905PEP-2025315045. DOCX

Claims

Claims1 . Decoder (100, 100’, 200) for decoding a signal, wherein the decoder is configured to decode an encoded representation (101) of a precision reduced spectral domain representation of a signal, in order to obtain a decoded version (211) of the precision reduced spectral domain representation, wherein the decoder is configured to decode an encoded representation (102) of a residual information, in order to obtain a decoded version (121) of the residual information, wherein the decoder is configured to obtain a filtered and / or prediction processed precision reduced spectral domain representation (131) on the basis of the precision reduced spectral domain representation using a filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and / or using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and wherein the decoder is configured to determine a combined spectrum (141) based on the filtered and / or prediction processed precision reduced spectral domain representation and the decoded version of the residual information.

2. Decoder (100, 100’, 200) according to claim 1 , wherein the decoded version (121) of the residual information represents a less significant portion of a number representation of a spectral domain representation of the audio signal, wherein the filtered and / or prediction processed precision reduced spectral domain representation (131) represents a more significant portion of the number representation of the spectral domain representation of the audio signal, and wherein the decoder is configured to combine the filtered and / or prediction processed precision reduced spectral domain representation and the decoded version of theFH240905PEP-2025315045. DOCXresidual information, in order to obtain the number representation of the spectral domain representation of the audio signal.

3. Decoder (100, 100’, 200) according to claim 2, wherein the decoder is configured to obtain a first number representation, which represents the filtered and / or prediction processed precision reduced spectral domain representation (131), wherein the decoder is configured to obtain a second number representation, which represents the decoded version (121) of the residual information, and wherein the decoder is configured to combine the first number representation and the second number representation, in order to obtain the number representation of the spectral domain representation of the audio signal.

4. Decoder (100, 100’, 200) according to claim 3, wherein the decoder is configured to concatenate the first number representation and the second number representation, in order to obtain the number representation of the spectral domain representation of the audio signal.

5. Decoder (100, 100’, 200) according to any of claims 2 to 4, wherein the decoded version (121) of the residual information represents one or more least significant bits, LSBs, of the number representation of the spectral domain representation of the audio signal, wherein the filtered and / or prediction processed precision reduced spectral domain representation (131) represents one or more most significant bits, MSBs, of the number representation of the spectral domain representation of the audio signal, and wherein the decoder is configured to combine the one or more LSBs and the one or more MSBs, in order to obtain the number representation of the spectral domain representation of the audio signal.

6. Decoder (100, 100’, 200) according to any of claims 2 to 5,FH240905PEP-2025315045. DOCXwherein the decoder is configured to determine the decoded version (121) of the residual information with variable accuracy, depending on an amount of residual information included in the encoded representation (102) of the residual information.

7. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoded version (121) of the residual information comprises a residual spectral domain representation, representing a difference between the spectral domain representation and the filtered and / or prediction processed precision reduced spectral domain representation (131).

8. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (131) using a prediction filtering in a frequency direction.

9. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (131) using a temporal noise shaping filter structure, in order to effect a prediction filtering in a frequency direction of the precision reduced spectral domain representation.

10. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (131) using a filter structure and / or using a prediction processing structure (130), wherein the decoder is configured to decode a set of filter parameters for use with the filter structure, and / or wherein the decoder is configured to decode a set of prediction processing parameters for use with the prediction processing structure.FH240905PEP-2025315045. DOCX11. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to decode filter parameters and / or prediction processing parameters (211), and wherein the decoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (131) using a filter structure and / or using a prediction processing structure (130), wherein the decoder is configured to adapt the filtering performed using the filter structure based on the filter parameters, in order to adapt a weighting of a prediction of the filtering, and / or to adapt a prediction order of the filtering and / or in order to reconfigure the filter structure; and / or wherein the decoder is configured to adapt the prediction processing performed using the prediction processing structure based on the prediction processing parameters, in order to adapt a weighting of the prediction processing, and / or to adapt a prediction order of the prediction processing and / or in order to reconfigure the prediction processing structure.

12. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to decode filter parameters and / or prediction processing parameters (211), and wherein the decoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain (131) representation using a filter structure and / or using a prediction processing structure (130), wherein the decoder is configured to adapt based on the filter and / or prediction processing parameters whether a filtering and / or a prediction processing is performed in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and / or wherein the decoder is configured to adapt for at least one filtering direction and / or for at least one combination of filtering directions the filtering performed using the filter structure based on the filter parameters, in order to adapt a weighting of a prediction ofFH240905PEP-2025315045. DOCXthe filtering, and / or to a adapt a prediction order of the filtering and / or in order to reconfigure the filter structure and / or in order to change filter coefficients of the filter structure; and / or wherein the decoder is configured to adapt for at least one prediction processing direction and / or for at least one combination of prediction processing directions the prediction processing performed using the prediction processing structure based on the prediction processing parameters, in order to adapt a weighting of the prediction processing, and / or to adapt a prediction order of the prediction processing and / or in order to reconfigure the prediction processing structure and / or in order to change a prediction coefficients of the prediction processing structure.

13. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoded version (121) of the precision reduced spectral domain representation comprises integer-valued spectral bin values for a plurality of spectral bins, and wherein the decoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (131) in the form of an integer valued filtered and / or prediction processed precision reduced spectral domain representation.

14. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoded version (121) of the precision reduced spectral domain representation comprises integer-valued spectral bin values for a plurality of spectral bins, wherein the decoder is configured to map integer values of the decoded version (111) of the precision reduced spectral domain representation onto integer values of the filtered and / or prediction processed precision reduced spectral domain representation (131) using a rounding of predicted and / or filtered and / or prediction filtered and / or prediction processed values.

15. Decoder (100, 100’, 200) according to any of the preceding claims,FH240905PEP-2025315045. DOCXwherein the decoded version (111) of the precision reduced spectral domain representation of a signal is an energy-reduced version and / or a redundancy-reduced version and / or a version with removed predictable portions of a precision reduced spectral domain representation of the signal, and wherein the decoder is configured to losslessly reconstruct the precision reduced spectral domain representation of the signal using the filtering and / or using the prediction processing.

16. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (131) on the basis of the precision reduced spectral domain representation (111) using a filtering which is independent of the residual information (121), and / or using a prediction processing which is independent of the residual information (121).

17. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to decode a filter mode information (443, 511) and / or a prediction processing mode information (443, 511), and wherein the decoder is configured to adapt the filtering of the precision reduced spectral domain representation (111) in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation (131) based on the filter mode information; and / or wherein the decoder is configured to adapt the prediction processing of the precision reduced spectral domain representation (111) in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation (131) based on the prediction processing mode information.

18. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to decode an information describing the operation mode of the encoder used to encode the encoded representation of a precision reduced spectral domain representation; andFH240905PEP-2025315045. DOCXwherein the decoder is configured to adapt the filtering and / or prediction processing of the precision reduced spectral domain representation (111) in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation (131) based on the information describing the operating mode of the encoder, and / or wherein the decoder is configured to selectively determine the combined spectrum (141) based on the filtered and / or prediction processed precision reduced spectral domain representation (131) and the decoded version (121) of the residual information or based on the filtered and / or prediction processed precision reduced spectral domain representation (131) without considering a residual information.

19. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to use a default residual information (242) instead of the decoded version (121) of the residual information, if the encoded representation (102) of a residual information is absent in a respective bitstream.

20. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to selectively enable or disable usage of a rounding functionality in the filtering in the frequency direction and / or in the time direction, and / or in the spatial direction and / or in the channel direction; and / or selectively enable or disable usage of a rounding functionality in the prediction processing in the frequency direction and / or in the time direction, and / or in the spatial direction and / or in the channel direction.

21. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the filtering in the frequency direction and / or in the time direction, and / or in the spatial direction and / or in the channel direction is configured to perform a lossless reconstruction of the filtered and / or prediction processed precision reduced spectral domain representation (131); and / orFH240905PEP-2025315045. DOCXwherein the prediction processing in the frequency direction and / or in the time direction, and / or in the spatial direction and / or in the channel direction is configured to perform a lossless reconstruction of the filtered and / or prediction processed precision reduced spectral domain representation (131).

22. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the filtering in the frequency direction is a TNS filter with lossless reconstruction.

23. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to receive an information describing a splitting between the representation of the precision reduced spectral domain representation of the signal and the representation of the residual information.

24. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to receive an information describing whether a rounding should be applied in the filtering in the frequency direction and / or in the time direction, and / or in the spatial direction and / or in the channel direction; and / or wherein the decoder is configured to receive an information describing whether a rounding should be applied in the prediction processing in the frequency direction and / or in the time direction, and / or in the spatial direction and / or in the channel direction.

25. Decoder (100, 100’, 200) according to any of the preceding claims, wherein the decoder is configured to switch between a lossless mode, a near lossless mode and a lossy mode; wherein the decoder is configured to decode and to losslessly filter and / or predict a spectral domain representation, if the decoder is in the lossless mode,FH240905PEP-2025315045. DOCXwherein the decoder is configured to decode and to losslessly filter and / or predict the precision reduced spectral domain representation (111), if the encoder is in the near lossless mode, and wherein the decoder is configured to decode and to lossily filter and / or predict a spectral domain representation, if the encoder is in the lossy mode.

26. Encoder (300, 300’, 400) for encoding a signal, wherein the encoder is configured to obtain a spectral domain representation of the signal (311); wherein the encoder is configured to obtain a precision reduced spectral domain representation of the signal (321) based on the spectral domain representation, wherein the encoder is configured to obtain a residual information (331) based on the spectral domain representation; wherein the encoder is configured to obtain a filtered and / or prediction processed precision reduced spectral domain representation (341) on the basis of the precision reduced spectral domain representation using a filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and / or using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, wherein the encoder is configured to obtain an encoded representation (351) on the basis of the filtered and / or prediction processed precision reduced spectral domain representation and the residual information.

27. Encoder (300, 300’, 400) according to claim 26, wherein the spectral domain representation (311) is an integer spectral domain representation of the signal,FH240905PEP-2025315045. DOCXwherein the encoder is configured to obtain the precision reduced spectral domain representation of the signal (321) in the form of an integer precision reduced spectral domain representation based on the integer spectral domain representation, and wherein the encoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341) as a filtered and / or prediction processed integer precision reduced spectral domain representation on the basis of the integer precision reduced spectral domain representation using a filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and / or using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction.

28. Encoder (300, 300’, 400) according to claim 27, wherein the encoder is configured to obtain the filtered and / or prediction processed integer precision reduced spectral domain representation (341) using a rounding operation.

29. Encoder (300, 300’, 400) according to any of claims 26 to 28, wherein the encoder is configured to obtain the precision reduced spectral domain representation (321) in the form of a more significant portion of a number representation of the spectral domain representation (311) of the audio signal.

30. Encoder (300, 300’, 400) according to claim 29, wherein the encoder is configured to obtain the precision reduced spectral domain representation (321) in the form of a first number representation.

31. Encoder (300, 300’, 400) according to claim 29 or 30, wherein the encoder is configured to obtain the precision reduced spectral domain representation (321) in the form of most significant bits, MSBs, of the number representation of the spectral domain representation of the audio signal.

32. Encoder (300, 300’, 400) according to any of claims 26 to 31 ,FH240905PEP-2025315045. DOCXwherein the encoder is configured to obtain the residual information (331) in the form of a residual spectral domain representation or a portion thereof based on the spectral domain representation (311).

33. Encoder (300, 300’, 400) according to any of claims 26 to 32, wherein the encoder is configured to obtain the residual information (331) in the form of a less significant portion of a number representation of a spectral domain representation (311).

34. Encoder (300, 300’, 400) according to any of claims 26 to 33, wherein the encoder is configured to obtain the residual information (331) in the form of a second number representation based on the spectral domain representation (311).

35. Encoder (300, 300’, 400) according to any of claims 26 to 34, wherein the encoder is configured to obtain the residual information (331) in the form of one or more least significant bits, LSBs, of the number representation of the spectral domain representation of the audio signal (311),36. Encoder (300, 300’, 400) according to any of claims 26 to 35, wherein the encoder is configured to adapt the filtering in frequency direction and / or in time direction and / or in spatial direction and / or in channel direction based on the available bitrate; and / or wherein the encoder is configured to adapt the prediction processing in frequency direction and / or in time direction and / or in spatial direction and / or in channel direction based on the available bitrate.

37. Encoder (300, 300’, 400) according to any of claims 26 to 36, wherein the encoder is configured to selectively use a filtering with rounding or a filtering without rounding and / or to selectively use a prediction processing with rounding or aFH240905PEP-2025315045. DOCXprediction processing without rounding, in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341).

38. Encoder (300, 300’, 400) according to any of claims 26 to 37, wherein the encoder is configured to use the filtering with rounding and / or the prediction processing with rounding if an available bitrate is sufficient to losslessly encode the filtered and / or prediction processed precision reduced spectral domain representation (341) obtainable by filtering and / or by prediction processing with rounding, and wherein the encoder is configured to use the filtering without rounding and / or the prediction processing without rounding if the available bitrate is insufficient to losslessly encode the filtered and / or prediction processed precision reduced spectral domain representation (341) obtainable by filtering and / or by prediction processing with rounding.

39. Encoder (300, 300’, 400) according to any of claims 26 to 38, wherein the encoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341) on the basis of the precision reduced spectral domain representation (321) using a temporal noise shaping filter structure, in order to effect a prediction filtering in a frequency direction of the precision reduced spectral domain representation.

40. Encoder (300, 300’, 400) according to any of claims 26 to 39, wherein the encoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341) using a filter structure, wherein the encoder is configured to obtain a set of filter parameters for use with the filter structure; and / or wherein the encoder is configured to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341) using a prediction processing structure,FH240905PEP-2025315045. DOCXwherein the encoder is configured to obtain a set of prediction processing parameters for use with the prediction processing structure.

41. Encoder (300, 300’, 400) according to any of claims 26 to 40, wherein the encoder is configured to use a temporal noise shaping filtering with rounding, if an available bitrate is sufficient to losslessly encode the filtered and / or prediction processed precision reduced spectral domain representation (341) obtainable by the temporal noise shaping filtering with rounding, and wherein the encoder is configured to use the temporal noise shaping filtering without rounding, if the available bitrate is insufficient to losslessly encode the filtered and / or prediction processed precision reduced spectral domain representation (341) obtainable by the temporal noise shaping filtering with rounding.

42. Encoder (300, 300’, 400) according to any of claims 26 to 41 , wherein the encoder is configured to encode a filter mode information (443, 511) indicating whether a filtering with rounding or a filtering without rounding was performed in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341); and / or wherein the encoder is configured to encode a prediction processing mode information (443, 511) indicating whether a prediction processing with rounding or a prediction processing without rounding was performed in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341).

43. Encoder (300, 300’, 400) according to any of claims 26 to 42, wherein the encoder is configured to perform the filtering and / or the prediction processing to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341) independent from the residual information (331).

44. Encoder (300, 300’, 400) according to any of claims 26 to 43, wherein the encoder is configured to switch between different operating modes in dependence on an available bitrate.FH240905PEP-2025315045. DOCX45. Encoder (300, 300’, 400) according to any of claims 26 to 44, wherein the encoder is configured to switch between a lossless mode, a near lossless mode and a lossy mode in dependence on an available bitrate; wherein the encoder is configured to obtain and to encode a filtered and / or prediction processed version of the spectral domain representation (444), if the encoder is in the lossless mode wherein the encoder is configured to encode the filtered and / or prediction processed precision reduced spectral domain representation (341) or at least a part thereof, if the encoder is in the near lossless mode, and wherein the encoder is configured to obtain and to encode a filtered and / or prediction processed version of the spectral domain representation (444) or at least a part thereof, if the encoder is in the lossy mode.

46. Encoder (300, 300’, 400) according to any of claims 26 to 44, wherein the encoder is configured to switch between a lossless mode, a near lossless mode and a lossy mode in dependence on an available bitrate; wherein the encoder is configured to perform a filtering in frequency direction with rounding and / or a filtering in time direction with rounding and / or a filtering in spatial direction with rounding and / or a filtering in channel direction with rounding, and / or a prediction processing in frequency direction with rounding and / or a prediction processing in time direction with rounding and / or a prediction processing in spatial direction with rounding and / or a prediction processing in channel direction with rounding of the spectral domain representation in order to obtain a filtered and / or prediction processed version of the spectral domain representation (444) in the form of an integer representation and to encode the filtered and / or prediction processed version of the spectral domain representation, if the encoder is in the lossless mode, wherein the encoder is configured to perform a filtering in frequency direction with rounding and / or a filtering in time direction with rounding and / or a filtering in spatial direction with rounding and / or a filtering in channel direction with rounding, and / or aFH240905PEP-2025315045. DOCXprediction processing in frequency direction with rounding and / or a prediction processing in time direction with rounding and / or a prediction processing in spatial direction with rounding and / or a prediction processing in channel direction with rounding of the precision reduced spectral domain representation in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341) in the form of an integer representation and to encode the filtered and / or prediction processed precision reduced spectral domain representation, or at least a part thereof, if the encoder is in the near lossless mode, and wherein the encoder is configured to perform a filtering in frequency direction without rounding and / or a filtering in time direction without rounding and / or a filtering in spatial direction without rounding and / or a filtering in channel direction without rounding, and / or a prediction processing in frequency direction without rounding and / or a prediction processing in time direction without rounding and / or a prediction processing in spatial direction without rounding and / or a prediction processing in channel direction without rounding of the spectral domain representation in order to obtain a filtered and / or prediction processed version of the spectral domain representation (444) and to encode the filtered and / or prediction processed version of the spectral domain representation, or at least a part thereof, if the encoder is in the lossy mode.

47. Encoder (300, 300’, 400) according to one of claims 44 to 46, wherein the encoder is configured to encode an information (443) about the operating mode of the encoder.

48. Encoder (300, 300’, 400) according to any of claims 26 to 47, wherein the encoder is configured to determine the precision reduced spectral domain representation (321) in dependence on an available bitrate and / or in dependence on the spectral domain representation (311).

49. Encoder (300, 300’, 400) according to any of claims 26 to 48, wherein the encoder is configured to determine a bit representation of the filtered and / or prediction processed precision reduced spectral domain representation (341) and to losslessly encode said bit representation.FH240905PEP-2025315045. DOCX50. Encoder (300, 300’, 400) according to any of claims 27 to 49, wherein the encoder is configured to determine a bit representation of the residual information (331) and to selectively encode said bit representation, a portion of said bit representation or to selectively not encode said bit representation in dependence on an available bitrate.

51. Encoder (300, 300’, 400) according to any of claims 26 to 50, wherein the encoder is configured to encode the filtered and / or prediction processed precision reduced spectral domain representation (341) and the residual information (331) separately in a bitstream.

52. Encoder (300, 300’, 400) according to any of claims 26 to 51 , wherein the encoder comprises a dedicated functionality for encoding the residual information (331), wherein said functionality is configured to adapt a size of the residual information or an amount of residual information that is encoded.

53. Encoder (500, 600), for encoding a signal, wherein the encoder is configured to switch between a noise shaping operating mode, and a redundancy reduction operating mode; wherein the encoder is configured to obtain a spectral domain representation (311) of the signal; wherein the encoder is configured to obtain a filtered and / or prediction processed spectral domain representation (444) on the basis of the spectral domain representation (311) using a filter structure and / or using a prediction processing structure, and obtain an encoded representation (351”) on the basis of the filtered and / or prediction processed spectral domain representation, if the encoder is in the noise shaping operating mode;FH240905PEP-2025315045. DOCXwherein the encoder is configured to obtain a precision reduced spectral domain representation of the signal (321) based on the spectral domain representation, obtain a filtered and / or prediction processed precision reduced spectral domain representation (341) on the basis of the precision reduced spectral domain representation using the filter structure and / or using the prediction processing structure, and obtain an encoded representation (351”) on the basis of the filtered and / or prediction processed precision reduced spectral domain representation, if the encoder is in the redundancy reduction operating mode.

54. Encoder (500, 600) according to claim 53, wherein the encoder is configured to obtain the filtered and / or prediction processed spectral domain representation (444) in the noise shaping operating mode and the filtered and / or prediction processed precision reduced spectral domain representation (341) in the redundancy reduction operating mode using the same filter structure and filter parameters, which are independent of whether the encoder is in the noise shaping operating mode or in the redundancy reduction operating mode, and / or using the same prediction processing structure and prediction processing parameters, which are independent of whether the encoder is in the noise shaping operating mode or in the redundancy reduction operating mode.

55. Encoder (500, 600) according to one of claims 53 to 54, wherein the encoder is configured to obtain the filtered and / or prediction processed spectral domain representation (444) in the noise shaping operating mode and the filtered and / or prediction processed precision reduced spectral domain representation (341) in the redundancy reduction operating mode using the same filtering and / or using the same prediction processing.

56. Encoder (500, 600) according to one of claims 53 to 55,FH240905PEP-2025315045. DOCXwherein the encoder is configured to obtain a residual information (331) based on the spectral domain representation and to obtain an encoded representation on the basis of the filtered and / or prediction processed precision reduced spectral domain representation (341) and on the basis of the residual information (331), if the encoder is in the redundancy reduction operating mode.

57. Encoder (500, 600) according to any of claims 53 to 56, wherein the encoder is configured to switch between the operating modes in dependence on an available bitrate.

58. Encoder (500, 600) according to any of claims 53 to 57, wherein the encoder is configured to switch between a lossless mode, a near lossless mode and a lossy mode in dependence on an available bitrate; wherein the encoder is configured to obtain and to encode a filtered and / or prediction processed version (444) of the spectral domain representation, if the encoder is in the lossless mode, wherein the encoder is configured to encode the filtered and / or prediction processed precision reduced spectral domain representation (341) or at least a part thereof, if the encoder is in the near lossless mode, and wherein the encoder is configured to obtain and to encode a filtered and / or prediction processed version of the spectral domain representation (444) or at least a part thereof, if the encoder is in the lossy mode.

59. Encoder (500, 600) according to any of claims 53 to 58, wherein the encoder is configured to switch at least between a a lossy version of the noise shaping operating mode, a near lossless version of the noise shaping operating mode, a near lossless version of the redundancy reduction operating mode, and a lossless version of the redundancy reduction operating modeFH240905PEP-2025315045. DOCXin dependence on an available bitrate; wherein the encoder is configured to perform a filtering in frequency direction with rounding and / or a filtering in time direction with rounding and / or a filtering in spatial direction with rounding and / or a filtering in channel direction with rounding, and / or a prediction processing in frequency direction with rounding and / or a prediction processing in time direction with rounding and / or a prediction processing in spatial direction with rounding and / or a prediction processing in channel direction with rounding of the spectral domain representation in order to obtain a filtered and / or prediction processed version of the spectral domain representation (444) in the form of an integer representation and to encode the filtered and / or prediction processed version of the spectral domain representation, if the encoder is in the lossless version of the redundancy reduction operating mode, wherein the encoder is configured to perform the filtering in frequency direction with rounding and / or the filtering in time direction with rounding and / or the filtering in spatial direction with rounding and / or the filtering in channel direction with rounding, and / or the prediction processing in frequency direction with rounding and / or the prediction processing in time direction with rounding and / or the prediction processing in spatial direction with rounding and / or the prediction processing in channel direction with rounding of the precision reduced spectral domain representation, in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341) in the form of an integer representation and to encode the filtered and / or prediction processed precision reduced spectral domain representation, or at least a part thereof, if the encoder is in the near lossless version of the redundancy reduction operating mode, and wherein the encoder is configured to perform the filtering in frequency direction with rounding and / or the filtering in time direction with rounding and / or the filtering in spatial direction with rounding and / or the filtering in channel direction with rounding, and / or the prediction processing in frequency direction with rounding and / or the prediction processing in time direction with rounding and / or the prediction processing in spatial direction with rounding and / or the prediction processing in channel direction with rounding of the precision reduced spectral domain representation, in order to obtain the filtered and / or prediction processed precision reduced spectral domain representation (341) in the form of an integer representation and to encode a portion of the filtered and / or prediction processed precision reduced spectral domain representation but noFH240905PEP-2025315045. DOCXresidual information, if the encoder is in the near lossless version of the noise shaping operating mode, wherein the encoder is configured to perform a filtering in frequency direction without rounding and / or a filtering in time direction without rounding and / or a filtering in spatial direction without rounding and / or a filtering in channel direction without rounding, and / or a prediction processing in frequency direction without rounding and / or a prediction processing in time direction without rounding and / or a prediction processing in spatial direction without rounding and / or a prediction processing in channel direction without rounding of the spectral domain representation, in order to obtain a filtered and / or prediction processed version of the spectral domain representation (444) and to encode a portion of the filtered and / or prediction processed version of the spectral domain representation, if the encoder is in the lossy version of the noise shaping operating mode.

60. Encoder (500, 600) according to one of claims 53 to 59, wherein the encoder is configured to encode an information (511) about the operating mode of the encoder.

61. Method for decoding a signal, the method comprising: decoding an encoded representation (101) of a precision reduced spectral domain representation of a signal, in order to obtain a decoded version (111) of the precision reduced spectral domain representation, decoding an encoded representation (102) of a residual information, in order to obtain a decoded version (121) of the residual information, obtaining a filtered and / or prediction processed precision reduced spectral domain representation (131) on the basis of the precision reduced spectral domain representation using a filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and / or using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, andFH240905PEP-2025315045. DOCXdetermining a combined spectrum (141) based on the filtered and / or prediction processed precision reduced spectral domain representation and the decoded version of the residual information.

62. Method for encoding a signal, the method comprising: obtaining a spectral domain representation of the signal (311); obtaining a precision reduced spectral domain representation of the signal (321) based on the spectral domain representation, obtaining a residual information (331) based on the spectral domain representation; obtaining a filtered and / or prediction processed precision reduced spectral domain representation (341) on the basis of the precision reduced spectral domain representation using a filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and / or using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, obtaining an encoded representation (351) on the basis of the filtered and / or prediction processed precision reduced spectral domain representation and the residual information.

63. Method for encoding a signal, the method comprising: switching between a noise shaping operating mode, and a redundancy reduction operating mode; obtaining a spectral domain representation of the signal (311); wherein the method further comprises, in the noise shaping operating mode, obtaining a filtered and / or prediction processed spectral domain representation (444) on the basis of the spectral domain representation using a filter structure and / or using a prediction processing structure andFH240905PEP-2025315045. DOCXobtaining an encoded representation (351”) on the basis of the filtered and / or prediction processed spectral domain representation; wherein the method further comprises, in the redundancy reduction operating mode, obtaining a precision reduced spectral domain representation of the signal (321) based on the spectral domain representation, obtaining a filtered and / or prediction processed precision reduced spectral domain representation (341) on the basis of the precision reduced spectral domain representation using the filter structure and / or using the prediction processing structure, and obtaining an encoded representation (351”) on the basis of the filtered and / or prediction processed precision reduced spectral domain representation.

64. Computer program for performing the method according to one of claims 61 , 62 or 63, when the computer program runs on a computer.

65. Bitstream, comprising: an encoded representation (101) of a precision reduced spectral domain representation of a signal, and an information indicating whether a rounding should be applied in a filtering in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction that is to be to applied to a decoded representation of the precision reduced spectral domain representation in a decoder, and / or an information indicating whether a rounding should be applied in a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction that is to be to applied to a decoded representation of the precision reduced spectral domain representation in a decoder.

66. Bitstream, comprising:FH240905PEP-2025315045. DOCXan encoded representation (101) of a precision reduced spectral domain representation of a signal, and an encoded representation (102) of a residual information, and an information describing a splitting between the representation of the precision reduced spectral domain representation of the signal and the representation of the residual information.

67. Decoder (100, 100’, 200) for decoding a signal, wherein the decoder is configured to decode an encoded representation (10T) of a prediction information, in order to obtain a decoded version (11T) of the prediction information, wherein the decoder is configured to decode an encoded representation (102) of a residual information, in order to obtain a decoded version (121) of the residual information, wherein the decoder is configured to obtain a prediction processed precision reduced spectral domain representation (131’) on the basis of the prediction information using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and wherein the decoder is configured to determine a combined spectrum (141’) based on the prediction processed precision reduced spectral domain representation and based on the decoded version of the residual information.

68. Encoder (300, 300’, 400) for encoding a signal, wherein the encoder is configured to obtain a spectral domain representation of the signal (311); wherein the encoder is configured to obtain a precision reduced spectral domain representation of the signal (321) based on the spectral domain representation,FH240905PEP-2025315045. DOCXwherein the encoder is configured to obtain a residual information (331) based on the spectral domain representation; wherein the encoder is configured to obtain a prediction information (34T) on the basis of the precision reduced spectral domain representation using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, wherein the encoder is configured to obtain an encoded representation (351’) on the basis of the prediction information and on the basis of the residual information.

69. Method for decoding a signal, the method comprising: decoding an encoded representation (10T) of a prediction information, in order to obtain a decoded version (11T) of the prediction information, decoding an encoded representation (102) of a residual information, in order to obtain a decoded version (121) of the residual information, obtaining a prediction processed precision reduced spectral domain representation (13T) on the basis of the prediction information using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, and determining a combined spectrum (14T) based on the prediction processed precision reduced spectral domain representation and based on the decoded version of the residual information.

70. Method for encoding a signal, the method comprising: obtaining a spectral domain representation of the signal (311); obtaining a precision reduced spectral domain representation of the signal (321) based on the spectral domain representation, obtaining a residual information (331) based on the spectral domain representation;FH240905PEP-2025315045. DOCXobtaining a prediction information (34T) on the basis of the precision reduced spectral domain representation using a prediction processing in a frequency direction and / or in a time direction and / or in a spatial direction and / or in a channel direction, obtaining an encoded representation (35T) on the basis of the prediction information and on the basis of the residual information.

71. Computer program for performing the method according to one of claims 69 or 70, when the computer program runs on a computer.

72. Bitstream, comprising: an encoded representation (10T) of a prediction information, representing a prediction processed version of a precision reduced spectral domain representation of a signal; and an encoded representation (102’) of a residual information, representing a residual between the precision reduced spectral domain representation of the signal and the spectral domain representation of the signal.FH240905PEP-2025315045. DOCX

Images